Optimized Railway Track Condition Monitoring and Derailment Prevention System Supported by Cloud Technology

Abstract

This paper describes an easy way to monitor railway track abnormalities and update information on the track’s status to the cloud. Abnormalities present in railway tracks should be identified promptly and rectified to ensure safe and smooth travel. In this paper, a cloud-based track monitoring system (CTMS) is proposed for the monitoring of track conditions. The micro-electro mechanical systems (MEMS) accelerometers which are mounted in the axle are used to measure the railway track abnormality. The measured signal is optimized using the flower pollination optimization algorithm (FPOA). Because of signaling problems in the global positioning system (GPS), it is difficult to estimate the exact location of the abnormality in real time. A new method is introduced to overcome this problem. It provides the location of an abnormality even when the GPS signal is absent. The performance of the CTMS is compared with three different speed scenarios of the vehicle. The information about the abnormality on the track can be shared with other trains that pass through the same location so that the driver can reduce speed in that location to avoid derailment. Finally, an experimental setup was developed and the performance of CTMS is studied under four different irregularity cases.

Rail transport plays a vital role in moving passengers and goods. Over the past century, the infrastructure of the rail industry has seen remarkable growth, particularly in developing countries. A railway track monitoring system was introduced by Weston et al. for detecting any abnormal conditions of the railway track ( 1 ). The in-service monitoring system computes the track parameters and informs the central server of the track’s status. Tsunashima et al. have done useful work on an in-service vehicle with a sensor arrangement. Their differential global positioning system (DGPS) may serve better in the detection of the abnormalities present on the track ( 2 ). Rail and track interactions generate a significant force on the railway track during service. Such force causes substantial vibration, reduction of safety, and degradation of the rail track ( 3 ). Numerous approaches have been developed by Weston et al. to estimate the condition of the track. The acceleration signal is used for predicting any abnormal vibration on the railway ( 4 ). Molodova et al. introduced a new method for detecting track surface problems. In this method, sensors are mounted in the axle box, and the track abnormalities are estimated from the acceleration signal ( 5 ). Primary cracks created by the rail–wheel dynamic force was studied by Deng et al. Two different shapes of squat defect were chosen and the problem diagnosed ( 6 ).

A precise test unit was developed by Zhai et al. to measure the health of the track for high-speed trains. In this method, both the wheel profile and the acceleration signal from the axle box are considered by an onboard measuring system ( 7 ). Continuous monitoring of the curved track using the probabilistic assessment method is achieved by large dynamic range and high sensitivity ( 8 ). The quality of the track has been estimated through the use of rail-wheel dynamics of the track. This is a useful work in the estimation of track abnormality caused by high-speed trains. Three different filters—bandpass, Kalman, and compensation—are used for estimation ( 9 , 10 ). An integrated track monitoring system of the various components used in the rail infrastructure system and its operation were studied and analyzed by Stenström et al. ( 11 ). Fuzzy logic-based track condition monitoring was used in the estimation of the abnormality of the track by Chellaswamy et al. ( 12 ).

Apart from a scholarly study, the commercial method for monitoring track conditions and its results should be cited. Non-stationary devices are used to monitor railway track conditions and are discussed in ( 13 ). These devices are used to monitor the track under operation and are used specifically on high-speed railway tracks. The track monitoring machine installed as an additional device for recording the status of the track is discussed in ( 14 ). Track monitoring of high-speed trains is achieved through the rail-track coupled dynamics model. The bearing stiffness of the axle box, mesh stiffness, and bearing clearance have been included in the model ( 15 ). An automatic system is used to monitor track abnormality. This generates a report, which is sent to the maintenance department so that necessary action can be discussed ( 16 ). A simulation study and an experimental test were carried out by Allotta et al. to estimate the abnormality of the track using a stationary device ( 17 ).

The Markov model is used to inspect different parts of the railway track such as twists, unevenness, and gauge measurement ( 18 ). The Markov model-based rail breakage probability was developed by Podofillini et al. to estimate the cost and risk associated with different checking strategies ( 19 ). Prescott and Andrews proposed a Markov model for analyzing the track degradation problem. The test was conducted on the UK rail network ( 20 ). In recent years, optimization techniques have inspired great interest in track monitoring. A Genetic Algorithm (GA) based structure vibration monitoring system was developed by Parhi et al. for smart buildings ( 21 ). And Particle Swarm Optimization (PSO) based parameter identification for a structural system was proposed by Xue et al. ( 22 ).

Internet of things (IoT) based railway track condition monitoring was proposed by Xiukun et al. The controller always watches the status of the track when the vibration exceeds the set threshold value and updates the corresponding coordinates (latitude and longitude) in the cloud ( 23 ). A video surveillance system for fog computation was proposed by Chen et al. and abnormal places are updated in the cloud ( 24 ). Broadband connectivity is needed for the continuous streaming of the video surveillance system. Accuracy of detection and quality in recognition is difficult for the cloud ( 25 , 26 ). Cloud-based technology has been used in different applications, including home applications (energy consumption of home appliances), industrial applications (online monitoring systems for continuous steel casting), medical applications, and so forth. ( 27 , 28 ). An IoT-based real-time networking was suggested by Huang et al. for humanoid robots. The efficiency, flow control, and priority were studied and compared with those of the existing network ( 29 ).

Train derailment can be avoided by continuous monitoring of the railway track. A track may develop problems for many reasons. Details of train derailments because of railway track problems at various locations in 2016 are given in ( 30 ). Among all derailments, the Pukhrayan train crash had the worst consequences: 150 people were killed, and a further 150 were injured. The damage present on the railway track should be recognized early and any failure to do this could result in lives being lost ( 31 ). A summary of the different methods used for monitoring the health of the railway track is given in Table 1. Even though these different methods have been used to monitor railway track conditions, no inspecting methods have provided simultaneous access to the location of abnormalities. The main focus of CTMS is to monitor the abnormality of the track and update the exact location in the cloud server and then share the location of the abnormality with other vehicles that will be passing through. The controller captures the location from the GPS when the acceleration signal exceeds the set threshold value. When a train passes through forests and hill areas where GPS and communication signals are absent or weak, the controller cannot identify the exact location of the abnormality. This problem has been solved through a new sensing method using GPS. This method automatically estimates the location of vibration even in the absence of a GPS signal. The proposed system also updates the irregularity of the track in the database and provides it to other trains passing along the same track. The train controller receives the information from the database and informs the driver early so that the speed of the train can be reduced and derailment avoided. The information can then be shared with the maintenance department (the authorities) so that they can take the necessary steps.

Table 1.

Summary of Different Methods Used to Monitor the Health of the Rail Track

Method	Author/year	Remarks
In-service vehicle measuring system	Xiukun et al. ( 23 )	The acceleration signal is received from the accelerometers mounted in the car body and bogie. The received signals are investigated using the frequency response analysis.A field test was conducted to verify consistency and accuracy.
Gyro sensor with map matching	Weston et al. ( 1 )	The yaw rate gyro is used to estimate the track irregularity.Four different trials were done, and a comparison is performed between the accelerometer and the gyro.The absolute location of irregularity is estimated through GPS, and map matching is used when the GPS signal is absent.
Wavelet spectrum analysis	Molodova et al. ( 5 )	Accelerometers are mounted in the axle box to estimate the squat defect of the rail track.It provides 100% accuracy in moderate and severe squat defects, and 78% small surface defects.
Fuzzy logic	Chellaswamy et al. ( 12 )	Fuzzy logic-based rail track condition monitoring was carried out using the acceleration signals received from the bogie and axle box.A comparison was performed between the conventional track measurement systems with the fuzzy logic based system under three different speed scenarios. The fuzzy logic based system provides better results.
Digital filtering	Kobayashi et al. ( 32 )	A track geometry recording was used to monitor the health of the rail track. The irregularity exceeds the predetermined value. The irregularity information with location information is sent to the control room.A comparison was performed between the digital inertial measuring devices with conventional measurements. The result shows the better processing speed and accuracy of the system.
Track recording method	Xiukun et al. ( 23 )	Track alignment is estimated through the data received from the acceleration sensors which are mounted in the bogie and car body of the train.A comparison was performed with the track recording vehicle, and it provides a good agreement between the measured and the recorded signals.
Vision-based system	Arun et al. ( 32 )	A vision-based rail track monitoring has carried out using drones. The analysis was performed with the drone image, and the result shows that reliability and accuracy.
ZigBee based railway track monitoring	Gao et al. ( 31 )	Wireless sensor nodes using ZigBee was used to monitor the railway track condition. It has a self-powered mechanism that includes three wireless nodes.Comparison was performed between two different tracks, and it provides an accuracy of 93.27%.
Neuro-Fuzzy Network	Mohammad and Mehdi ( 33 )	Fuzzy logic and neural networks were used to detect faults in the rail track. The audio frequency track circuit was used to measure the signal and train the Fuzzy-Neuro system.

Note: GPS = global positioning system.

In this study, four different cases of abnormalities—visible and invisible abnormalities, plastic deformation, and rust deformation on the top of the railway tracks—were examined and are shown in Figure 1. The innovations introduced by the CTMS are as follows:

FPOA algorithm was used to optimize the signal received from the MEMS accelerometer.

The location of the track abnormality estimation system was introduced to enable identification of the fault location with accuracy when the GPS signal is absent.

Abnormality information is shared with other trains that will be coming through the same location.

The rest of this paper is organized as follows. The next section provides an overview of the workflow of the proposed CTMS. We then describe the problems in signal coverage of GPS. In the subsequent section, we describe the simulation and experimental results. We end with our conclusions.

Figure 1.

Examples of track abnormality cases: (a) visible abnormality (missing bolt), (b) invisible abnormality (loose bolts and battered rail surface), (c) deformation with crack, and (d) rusty deformation on rail top.

Proposed System

In this section, the cloud-based track monitoring framework for avoiding train derailment is discussed. The cloud system integrates with different kinds of fixed and mobile sensors for storage, processing, and networking. The CTMS can be installed in passenger trains, goods trains, or both, and located in a goods wagon, a locomotive, a passenger car, a railcar, or other rail vehicles for the detection of the location of any abnormality. MEMS accelerometers were mounted in the vertical and lateral directions of the axel box of the rail vehicle. The signal produced by the MEMs is conditioned and sampled with a sampling frequency of 2 kHz. The analog-to-digital converter is used to convert the sampled signal into a digital signal before it is passed on to the controller after filtering. The FPOA optimization method is used to enhance the accuracy of detection.

The controller continues to track the signal levels of the GPS and the sensors and sends the geo-location data to the OpenGTS server when an abnormality is detected. A MongoDB distributed database was introduced to avoid problems arising from unstructured geo-location data that may be sent from the controller ( 34 , 35 ). When an abnormality is detected, an alert message is sent to the driver’s dashboard at least two kilometers before the abnormality (this alert distance can be changed through suitable programming) to help the driver to manage speed in the specified location.

Four different abnormalities (faults)—visible damage (missing bolt), invisible damage (loose bolts and battered rail surface), deformation through cracks, and rusty deformation on railway tops—are considered and the study locations were in the Coimbatore area. The structure of the track monitoring is shown in Figure 2. The tracker device (DGPS-DSM132) was installed in the vehicle to identify the exact location of the abnormality. Figure 3 shows the tracking device to measure geo-location (latitude and longitude). Since it is accurate, compact, and easy to handle, it can be used in tracking vehicles or other mobile objects. The signal problem in GPS is explained below.

Figure 2.

Proposed architecture for smart track monitoring system.

Figure 3.

Differential global positioning system tracking device DSM132 for location measurement.

Furthermore, data on the abnormality’s coordinates can be transferred from the rail vehicle to an OpenGTS server through a 4G network. Such systems are currently used in tracking public vehicles such as ATM buses in Messina (The OpenGTS project, http://opengts.sourceforge.net). Web-based abnormality tracking can be done using OpenGTS (Open-Source GPS Tracking System). Here, the communication between trackers and OpenGTS is performed using a Restful approach. Spread diffusion technology offers the best application for trains such as a locomotives, high-speed trains, goods wagons, and so forth.

The information is gathered using two systems: (1) OpenGTS which visualizes the geo-located data on a map using OpenStreetMap (The OpenDMTP Project. http://www.opendmtp.org and http://opengts.sourceforge.net/documentation.html). This visualized information provides a clear picture of the location of the abnormality to help the driver reduce speed and avoid derailment; and (2) GeoJSON, an open standard format representing simple geographic data structures and their attributes. Since OpenGTS stores data inside an SQL database, a SQL-GeoJSON translation is needed to visualize the geo-information.

Accelerometers have been used to generate signals based on abnormal track conditions. The controller receives the signal from the accelerometer and checks whether its values are within the set threshold. The parameters of the accelerometer are presented in Table 2. The FPOA algorithm was used to optimize the data measured from the accelerometer. Optimization minimizes the difference between the estimated and the measured values. The various steps involved in this process are:

Table 2.

Technical Specifications of ADXL335Z Accelerometer

Parameters	Values
TypeImage	ADXL335Z
Manufacturer	Analog Devices
Axis	3
Range (x and y axis)For z axis	0.5–00 Hz0.5–550 Hz
Voltage (V)	1.8–3.6
Current (µA)	350
Interface	SPI, I²C
Precision	±3 g
Temperature (°C)	85
Price ($)	3

a set of real-time data is measured from the accelerometer;

an objective function is defined for minimizing the difference between real and estimated values;

the parameter values are tuned by applying FPOA until the best value is obtained; and

the optimal value is extracted from the obtained solution.

Workflow of the Proposed System

The workflow, presented in Figure 4, provides the solution for the track monitoring and information system. Initially, it begins with the time series model and estimates the abnormality present in the track with information related to the location.

Figure 4.

Workflow of the algorithm for track monitoring system; smart cloud-based track monitoring system automatically estimates the abnormality of the track and perform the required action.

The algorithm makes continuous checks on the status of the track and the controller updates it in the cloud if an abnormality is detected through GPS. The control information is sent to the driver before reaching a particular area. The CTMS consists of three main sections, namely, track measurement, the cloud server, and decision making. The track measurement section consists of the accelerometer signal (Ax), the DGPS data (D_DGPS), and the ESP8266 (D_IoT) for data transfer to the cloud. A proximity sensor (D_PRO) is incorporated to continuously monitor the geo-location. If DGPS is absent or less than the set threshold value (D_THD), the controller switches on the proximity and estimates the distance to the abnormality by summing the D_DGPS and D_PRO (the distance measured by the proximity is explained further below) and updates the information to the cloud through ESP8266. If D_IoT is absent, then it will temporarily store and forward the information when the signal is available.

Duplication of data is created when different vehicles move through the same location of an abnormality. It can be avoided by the controller of CTMS, by comparing the received fault information (D_rec) to the current one (D_cur). If D_rec and D_cur are equal, it does not update further even on the detection of any abnormality in the same location and assumes the data are duplicated and discards them. The location of the abnormality is communicated from the cloud server to trains that will be traveling through that location. When a train receives details of the location of the fault (D_vib), the controller immediately alerts the driver with sound and visual information.

FPOA Algorithm

A Flower Pollination Optimization Algorithm (FPOA) is proposed by Chellaswamy et al. ( 36 ). Transfer of pollen from one species to another is known as pollination; this can be either by self-pollination (Abiotic) or by cross-pollination (Biotic). In practice, 10% of pollination is self-pollination, and the remaining 90% is cross-pollination. The control between self and cross-pollination is restricted through the probability P∈[0, 1]. The reproduction probability is used to measure the flower constancy. This algorithm has several distinguished merits; it has only limited complexity, it has the ability to explore and exploit the search space, it achieves faster convergence, it requires less effort for parameter tuning, and it is easy to program and compile. These significant advantages of the FPOA method motivate the authors to adopt this method for constructing reliable, fast, and rugged CTMS. The following rules are used to implement FPOA:

Rule 1: Cross-pollination follows levy flight for transfer of pollens globally. For example, the mth pollen on nth iteration for cross-pollination can be expressed based on ( 36 ) as:

P_{m}^{n + 1} = P_{m}^{n} + K (gbest - P_{m}^{n})

(1)

where K is the levy factor responsible for the transfer of pollens, and G_best represents the best solution obtained in $P_{m}^{n}$ , set of pollens. Since Levy distribution is used to transfer the pollens, the flight of pollen can be expressed based on ( 36 ) as:

K = \frac{μ Γ (μ) \sin (\frac{π μ}{2})}{π} \frac{1}{R^{1 + μ}}, (R >> R^{0} > 0)

(2)

where $Γ (μ)$ is the gamma function with the distribution range ( $R >> R^{0} > 0$ ).

Rule 2: Self-pollination is characterized by the local pollination process and it can be expressed as:

P_{m}^{n + 1} = P_{m}^{n} + β (P_{i}^{n} - P_{j}^{n})

(3)

where $P_{i}^{n} and P_{j}^{n}$ represent the flowers of the same species but different pollens. The probability switch S_p∈[0, 1] is used to control the switching between pollination. Moreover, this method is best suited for railway track condition monitoring because it introduces randomness in every iteration through self-pollination.

FPOA Implemented for CTMS Application

The various steps that would be involved in implementing FPOA in a CTMS application are devised and explained in the following steps:

Step 1. Parameter initialization: Maximum iteration number (I_N), probability switch (S_p), population size of initial duty cycle (P₁, P₂, P₃, P₄, and P₅), and limits on duty cycle (X_min and X_max) are set to 20, 0.15, 0.9, 0.6, and 7, respectively. Figure 5a shows the parameter initialization of FPOA method.

Step 2. Fitness evaluation: The stability of the pollens is evaluated by using the defined fitness function. The change of position of pollens indicates the evaluated value by the fitness function. The size of the stars in Figure 5b shows the best fitness value.

Step 3. Processing of pollination: In Figure 5b, the pollen (P₄) has higher fitness value, marked as G_best. Every pollen in the group must undergo either self or cross-pollination. A condition (if rand>P) is generated between the random number (0<rand<1) based on S_p.

Cross-pollination: In the present problem, one undergoes self-pollination and the remaining four pollens are cross-pollinated. The pollen (P₁) move toward GMPP through the vector (OA) with the levy factor, $K (gbest - A_{m}^{n})$ is shown in Figure 5c.

Figure 5.

Updation of pollens in flower pollination optimization algorithm algorithm: (a) initialization, (b) evaluation of fitness function, (c) cross-pollination, (d) self-pollination step, and (e) updation before the next iteration.

Self-pollination: In the first iteration, the pollen (P₃) undergoes local pollination. The next iteration update for the position of P₃ is shown in Figure 5d.

Step 4. New position of pollens: Based on Step 3, all pollens arrive at their new position via self, or cross-pollination as is shown in Figure 5e.

Step 5. Termination criterion: The steps between 2 and 5 are repeated until the maximum iteration is reached (I_N = 50) or the pollens get a maximum value.

The threshold values can be set based on the data set of the railway track. Based on the changes, the FPOA can reinitialize the search process from Step 1.

Signal Coverage Problem

In real life, a train may go through a forest area, tunnels, or mountains that cause fading of the signal without the ability for the controller to identify the exact location of vibration. Therefore, these problems are carefully considered for appropriate estimation of the fault location of the abnormality.

DGPS Coverage Problem

The DGPS module DSM132 was used in this study as it has higher accuracy and less tolerance up to 10 cm. In a real-time situation, a train might pass through forests or hill areas where the signal coverage of DGPS will be reduced making it more difficult for the system to estimate the exact location of any abnormalities (LOA) ( 37 ). In this study, a new location detection method is introduced; a proximity sensor is interfaced with the controller doing the estimation of the LOA even in the absence of the DGPS signal.

The signal coverage of DGPS is very important in estimating the LOA ( 37 ). Thus, the problem of signal coverage is studied in this paper. The coverage area of the DGPS and estimation of the LOA are illustrated in Figure 6. The proximity sensor is mounted in a wheel disc and produces one pulse per rotation. The controller counts the number of pulses generated by the proximity sensor and estimates the distance of the uncovered signal area of the GPS. Therefore, the controller estimates the distance (D) as:

D = n \times k

(4)

where n and k, denote, respectively, the number of pulses generated by the proximity sensor and the external circumference of the disc. For the estimate, the LOA based on Figure 6, is given by

LOA = Coordinat e_{old} + D

(5)

where Coordinate_old and D, denote, respectively, the point where the DGPS signal is lost and the distance traveled by the train from the Coordinate_old to the abnormality point. For instance, D = 6 km, then CTMS determines the place of vibration by just adding 6 km (the controller takes the direction toward the destination of the train) to the coordinates previously stored. Different states of signals and the corresponding operation performed by the controller are listed in Table 3. In the first case, the DGPS signal is lost so the CTMS automatically switches the proximity sensor on and estimates the location of vibration based on ( 5 ). Both the DGPS and communication signals are available in the second case; the controller sends the fault location of the vibration without any trouble. In the third case, the communication signal is absent, and DGPS is present. In this case, the controller stores the coordinates of the abnormal location and forwards it to the cloud database when the communication signal is available. In the final case, neither the DGPS or communication signals are available. When the controller senses this case, they switch the proximity on immediately and store the LOA if an irregularity is present, transferring the information later, when the communication signal is available.

Table 3.

Status of Global Positioning System (GPS), Communication Signal, Proximity Sensor, and the Controller Operation

DGPS signal	Communication signal	Controller operation	Condition of proximity
Not present	Present	Switch on the proximity, estimate the location of vibration based on ( 5 ) and send to the cloud if abnormality is detected	On
Present	Present	Send GPS coordinates if vibration present	Off
Present	Not present	Store GPS coordinates if vibration present	Off
Not present	Not present	Store the location of vibration and switch on the proximity and estimate location of any abnormalities if abnormality is detected	On

Figure 6.

Differential global positioning system coverage problem and estimation of LOA.

Results and Discussion

The analysis of the prototype of the proposed CTMS with different experiments is presented in this section. The objective of the experiment is to optimize the signal received from the track, estimating the exact location of an abnormality even when GPS signal is absent, and passing the information on the abnormality to other trains which will be passing through that particular area. It helps the driver to understand the problem in a specific location and reduce the speed to avoid derailment. The experiment was done for the four different cases: (1) visible damage (missing bolt), (2) rusty deformation on rail top, (3) invisible damage (loose bolts and battered rail surface), and (4) deformation with crack. The proposed CTMS method is compared with two other optimization methods, PSO and GA.

A prototype of CTMS was developed using the ARM processor (LPC2148) for railway track condition monitoring, and it was installed on the super-fast train (26142). The acceleration signal for different test cases with the corresponding coordinates was recorded under running conditions. The experimental setup of CTMS and the coordinates received for the corresponding location of an abnormality are shown in Figure 7. A large amount of data—around 78 GB—was stored on December 27, 2018. Two-way travel (departure and arrival) was considered for the four different abnormality locations around the Coimbatore area and the LOA was observed. The controller verified the fault in a particular position in both the departure and arrival of the same train. Identification of the abnormality in two-way journey in the same location confirmed the abnormality.

Figure 7.

Experimental setup of the proposed cloud-based track monitoring system.

Different field tests were carried out for all four abnormalities in the test location. Initially, the test was carried out under normal railway track conditions, and the data was stored under different speed scenarios. An abnormality was intentionally created for four different routes: Mettupalayam to Coimbatore; Wayalar to Chettipalayam; Coimbatore to Somanur; and Samattur to Valparai to study the performance of CTMS under three different speed scenarios. The abnormalities—the missing bolt (A), loose bolts (B), deformation with crack (C), and severe plastic deformation on track tops (D)—were chosen in the above places are shown in Figure 8.

Figure 8.

Test rail track at Coimbatore area.

The performance of the FPOA, CPSO, and GA was checked by different parameters such as size of population. The number of iterations is shown in Figure 9. Initially, the global minima increase rapidly and settle if population size increases. Figure 9a shows that the proposed FPOA settled around the population size of 53. On the other hand, the CPSO and GA are settled around the population size of 65 and 74 respectively. The effects of the number of iterations on the three algorithms are shown in Figure 9b. The global minima are settled around 24 iterations for FPOA, 45 iterations for CPSO, and 65 iterations for GA. From this study it is observed that the proposed FPOA provides better performance than CPSO and GA methods.

Figure 9.

Study of various parameters of flower pollination optimization algorithm, particle swarm optimization, and genetic algorithm: (a) size of population, and (b) number of iterations.

The convergence characteristics of FPOA, CPSO, and GA are shown in Figure 10. It is observed that the proposed FPOA converges faster than the other two methods: CPSO and GA. The abnormality signals received from the axle box accelerometer in the departure of the train and optimized using FPOA, CPSO, and GA for the four different cases are shown in Figure 11.

Figure 10.

Convergence characteristics of flower pollination optimization algorithm, particle swarm optimization, and genetic algorithm.

Figure 11.

Abnormality in axle box mounted accelerometer during departure: (a) invisible damage (loose bolts and battered rail surface), (b) visible damage (missing bolt), (c) deformation with crack, and (d) rusty deformation on rail top.

It is observed from Figure 11a that the values of displacement are higher at 6.15 km. The controller (LPC2148) always watches the received displacement level and compares it with the set threshold value. The threshold value was defined using different trials taken from the railway track, which is initially constructed in good condition. The controller stores the GPS coordinates (longitude and latitude) when it detects the abnormality (the received signal exceeds the threshold level). The visible damage reported by the accelerometer was optimized by FPOA, PSO, and GA, and is illustrated in Figure 11a. Figure 11a shows the presence of the abnormality at 6.2 km along the Mettupalayam to Coimbatore route. The invisible damage signal optimized using FPOA, PSO, and GA is shown in Figure 11b. The presence of the abnormality at 7 km from the start of the track on the Wayalar to Chettipalayam route was observed. Figure 11, a and b, reveals the higher signal levels produced by visible damage compared with invisible damage, and the location of the abnormality is identified with precision. The signal received from the accelerometer for deformation with crack is shown in Figure 11c. The graph illustrates the presence of the abnormality at 3.6 km from the start of the track on the Coimbatore to Somanur route. The irregularity of the rusty deformation on the top of the rail track is shown in Figure 11d. The graph shows the presence of the abnormality at 5.25 km on the Somanur to Valparai route.

The abnormality signal received from the accelerometer mounted in the axle box at the time of arrival in the same test location is shown in Figure 12. The experiment was carried out in all four cases. A comparison between Figures 11 and 12 shows that the visible abnormality, invisible abnormality, and plastic deformation on the top of the track were identified accurately. On the other hand, deformation with crack was identified in the departure time, but identification in the arrival time by FPOA was only slight but not by PSO and GA. So, the abnormality of the track was identified accurately by the proposed method compared with other methods.

Figure 12.

Abnormality in axle box mounted accelerometer during arrival: (a) invisible damage (loose bolts and battered rail surface), (b) visible damage (missing bolt), (c) deformation with crack and (d) rusty deformation on rail top.

Various tests were carried out in the test locations to check the performance of CTMS in identifying the location in both the departure and arrival of the train. When the signal level exceeded the set threshold, the controller identified the abnormality and captured the location from the DGPS. The experimental results of various track abnormalities for the proposed CTMS are shown in Figure 13. Figure 13a indicates the presence of the abnormality at coordinates 11.1891 N and 76.9446 E. A comparison of the GPS coordinates with the fault set coordinates shows that both of them almost matched with the fault coordinates. Figure 13b shows the recorded coordinates of invisible damage, indicating, by the sudden increase in data packet 43, the presence of an abnormality at coordinates 11.5394 N and 76.8986 E. The result shows the DGPS coordinates matching exactly with the fault set coordinates. Figure 13, c and d, shows the recorded coordinates of deformation with crack and rusty deformation on the track top respectively.

Figure 13.

Experimental results for different location of abnormalities with coordinates: (a) visible damage (missing bolt), (b) invisible damage (loose bolts and battered rail surface), (c) deformation with crack, and (d) rusty deformation on rail top.

In the third case, the GPS was switched off deliberately and the performance was tested. In this case, the controller immediately switched on the proximity and started estimating the distance. Table 4 shows the controller estimating the distance at 1.53 km and detecting the abnormality exactly at coordinates 12.9801 N, 80.2184 E. On the other hand, in the last case, the minute crack was not identified by the controller; a deviation of 0.94 m is present from the fault set coordinates. The same experiment was carried out for the arrival time of the train in the same test locations. The coordinates received match the departure coordinates exactly. The GPS signal level is forced to reduce in the minute crack case. The controller switched on the proximity and estimated the fault location during arrival.

Table 4.

Location of Abnormality and the Corresponding Global Positioning System Coordinates

Location of abnormality	Type of abnormality	Fault set coordinates		Fault identified coordinates		Distance estimated by proximity (km)	Accuracy (%)
		Latitude	Longitude	Latitude	Longitude
Departure
A	Visible Fault	13.0212° N	80.2231° E	13.0212°N	80.2231° E	No	100
B	Invisible Fault	13.0416° N	80.2349° E	13.0416° N	80.2349° E	No	100
C	Deformation with crack	12.9801° N	80.2184° E	12.9801° N	80.2184° E	1.53	99.17
D	Rusty deformation on the top	13.0065° N	80.2202° E	13.0065°N	80.2202° E	No	100
Arrival
A	Visible fault	13.0212° N	80.2231° E	13.0212° N	80.2231° E	No	100
B	Invisible fault	13.0416° N	80.2349° E	13.0416° N	80.2349° E	No	100
C	Deformation with crack	12.9801° N	80.2184° E	12.9801° N	80.2184° E	No	100
D	Rusty deformation on the top	13.0065° N	80.2202° E	13.0065° N	80.2202° E	2.75	100

OpenGTS is open-source and is designed to provide web-based tracking for vehicles under the Apache Software License. Moreover, this technology permits use by general-purpose vehicles, different kinds of satellite tracking systems, and private transport vehicles. It supports the visualization of the location of abnormalities on the track on an OpenStreetMap. The real-time track abnormality is shown in Figure 14. It allows monitoring of the abnormal location and storage of the data inside the MongoDB database. This unstructured data provides more flexibility for data analysis and manipulations.

Figure 14.

Visualization of OpenStreetMap for real-time track abnormality monitoring.

Statistical analysis was carried out for the evaluation of the results of the proposed CTMS, and the performance was compared with other related works. The analysis was carried out for the recorded data packet received from the accelerometer. The test was performed and the values recorded on December 27, 2018. Statistical metrics such as Root Mean Square Error (RMSE), Relative Error (RE), and Standard Deviation (SD) were used in this evaluation. These metrics can be expresses as:

RMSE = \sqrt{\frac{\sum_{k = 1}^{N} {(P_{mes, i} - P_{est})}^{2}}{N}}

(6)

RE = \frac{\sum_{k = 1}^{N} (P_{mes, i} - P_{est})}{P_{est}}

(7)

SD = \sqrt{\frac{\sum_{k = 1}^{N} {(P_{mes, i} - \bar{P_{mes, i}})}^{2}}{N}}

(8)

where

P_mes,i represents the measured signal power obtained using the optimization technique for the ith run,

P_est is the estimated power,

N represents the number of runs, and

$\bar{P_{mes, i}}$ is the average power measured using the optimization technique.

Various parameters used to analyze the performance of the proposed CTMS are: the population size = 10; the number of iterations = 20; and the number of executions = 50.

A comparison between the proposed CTMS and various other methods using different statistical metrics is shown in Table 5. There are three sets of samples and the corresponding RMSE, RE, and SD was calculated. Table 5 shows variations in the RMSE from 2.045 to 13.98, the lowest value for CTMS, and the highest value for ACO. The minimum RE of the proposed method was 2.037 followed by CPSO and GA. The SD is also less for the proposed method. The conclusion is that the proposed method is suitable for detecting track abnormalities compared with other methods while ACO has the lowest performance.

Table 5.

Satatistical Performance of Different Optimization Techniques

Abnormalities	Algorithm
	FPOA	CPSO	GA	ACO	FPOA	CPSO	GA	ACO
	RMSE				Relative error
Visible fault	2.045 2.312 2.163	3.329 3.134 3.628	3.637 3.572 4.271	3.888 3.946 4.103	2.127 2.037 2.293	2.038 2.381 2.728	2.733 2.836 3.281	2.978 2.826 3.182
Invisible fault	10.25 9.720 8.371	12.680 11.937 10.469	13.961 12.642 12.730	13.98 12.942 13.729	2.921 2.715 2.862	2.935 3.491 3.625	2.151 3.183 3.472	2.363 3.382 3.923
Deformation with crack	5.793 5.133 5.282	6.644 7.281 5.628	7.845 8.242 5.946	8.122 8.472 6.193	2.589 2.102 2.3.27	2.520 2.810 2.853	2.884 2.983 3.102	2.101 3.134 3.026
Rusty deformation on the top	6.724 7.182 6.299	9.264 7.292 6.406	9.833 7.829 6.628	10.28 7.972 6.925	2.472 3.281 2.738	2.689 3.528 2.892	2.983 3.498 2.910	2.994 3.790 2.978
Standard deviation
Visible fault	4.395 5.026 4.836	4.738 5.103 4.927	5.619 5.427 5.283	5.428 5.622 5.392
Invisible fault	16.26414.23115.683	15.38515.02715.729	16.58715.38015.820	16.48415.48315.839
Deformation with crack	8.3657.2847.502	8.7927.4937.538	9.6827.5287.583	10.7117.8507.729
Rusty deformation on the top	12.53010.36311.379	13.63310.52911.492	15.63010.70211.630	17.35210.82111.725

Note: FPOA = flower pollination optimization algorithm; CPSO = chaos particle swarm optimization; GA = genetic algorithm; ACO = ant colony optimization; RMSE = root mean square error.

In this study, network performance was studied by considering the constant latency of the network. The author has used data forwarded from the vehicle to the infrastructure and sent alert information from the infrastructure to the vehicle. The analysis is focused on inquiry and data processing, and data insertion and parsing. The proposed CTMS has two types of server; the first is used to store the data and the next for data collection and manipulation. The data received from the track was manipulated by the OpenGTS software installed in the i7-6700 CPU with 32 GB RAM. The MongoDB software is installed in the data storage server and configured under a single server mode.

Scalability analysis was done for both scenarios and, to obtain a reliable result, the test was repeated 40 times. The performance of parsing row data to GeoJSON and its inclusion in the MongoDB is shown in Figure 15. Here, data size and the corresponding response times were taken to plot the chart. The chart shows that the size of the data and response times are proportional, with a linear increase in the response time following an increase in data size.

Figure 15.

Performance comparison of parsing row data in MongoDB.

The author has considered 1,000 documents and agreed on the response time in this study of the performance of MongoDB. Variations in the distance were made from 1 km to 8 km, and the response time was studied. Details in Figure 16 show an increase in the distance, with slight variations in response time around 17 ms.

Figure 16.

Performance comparison of the proposed method in processing time.

Deliberate stress was made through an increase in the number of requests to test the scalable performance of the proposed CTMS. 1,200 queries were made, and the response is shown in Figure 17. In addition, Figure 17 illustrates the linear increase in response time following an increase in the number of requests.

Figure 17.

Performance comparison of the proposed method in query time.

Conclusion

Railway track condition monitoring is essential to control track vibration and avoid derailment. This paper presents a cloud-based track condition monitoring system which updates information on track abnormalities to the cloud. It enables identification of the abnormality location by the vehicle on the same track, enabling the driver to decrease speed in the specified sections of the route. MEMS accelerometers were incorporated in the axle box of the train and the abnormalities were tested under four abnormal cases. The conclusion from the test result was that the CTMS provides the exact LOA and updates it into the cloud to be shared with other vehicles. This helps avoid network complexity. The network performance parameters, such as processing speed and query time were studied. A prototype was developed and a different field test conducted, with the results showing the accuracy of the proposed system. The authors believe that CTMS is suitable for early detection of abnormalities on the track to avoid derailment.

Footnotes

Author Contributions

The authors confirm contribution to the paper as follows: study conception and design: C. Chellaswamy; data collection and analysis: T. S. Geetha; interpretation of results and editing: M. Surya Bhupal Rao; draft manuscript preparation: A. Vanathi. All authors reviewed the results and approved the final version of the manuscript.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

References

Weston

Roberts

Yeo

Stewart

Perspectives on Railway Track Geometry Condition Monitoring from In-Service Railway Vehicles. Vehicle System Dynamics, Vol. 53, No. 7, 2015, pp. 1063–1091.

Tsunashima

Mori

Yanagisawa

Ogino

Asano

Condition Monitoring of Railway Tracks Using Compact Size On-Board Monitoring Device. Proc., 6th IET Conference on Railway Condition Monitoring, Birmingham, 2014. https://doi.org/10.1049/cp.2014.0996.

Bogacz

Czyczula

Konowrocki

Influence of Sleepers Shape and Configuration on Track–Train Dynamics. Shock and Vibration, Vol. 2014, 2007. https://dx-doi-org.web.bisu.edu.cn/10.1155/2014/393867.

Westeon

P. F.

Ling

C. S.

Goodman

C. J.

Roberts

C. P.

Goodall

R. M.

Monitoring Lateral Track Irregularity from In-Service Railway Vehicles. Proceedings of the Institution of Mechanical Engineers Part F: Journal of Rail and Rapid Transit, Vol. 221, 2007, pp. 89–100.

Molodova

Núñez

Dollevoet

Automatic Detection of Squats in Railway Infrastructure. IEEE Transactions on Intelligent Transportation Systems, Vol. 15, 2014, pp. 1980–1990.

Deng

Qian

Dollevoet

Investigation of the Formation of Corrugation-Induced Rail Squats Based on Extensive Field Monitoring. International Journal of Fatigue, Vol. 112, 2018, pp. 94–105.

Zhai

Liu

Lin

Wang

Experimental Investigation on Vibration Behaviour of a CRH Train at Speed of 350 km/h. International Journal of Rail Transportation, Vol. 3, No. 1, 2015, pp. 1–16. https://doi.org/10.1080/23248378.2014.992819.

Zhai

Probabilistic Assessment of Railway Vehicle-Curved Track Systems Considering Track Random Irregularities. Vehicle System Dynamics, Vol. 56, No. 10, 2018, pp. 1552–1576.

Konop

Konowrocki

On Evaluation of the Wheelsets-Track Interaction Quality in Railway Engineering. Machine Dynamics Research, Vol. 37, 2013, pp. 61–70.

10.

Lee

J. S.

Choi

Kim

S. S.

Park

Kim

Y. G.

A Mixed Filtering Approach for Track Condition Monitoring Using Accelerometers on the Axle Box and Bogie. IEEE Transactions on Instrumentation and Measurement, Vol. 61, 2012, pp. 749–758.

11.

Stenström

Parida

Lundberg

Kumar

Development of an Integrity Index for Benchmarking and Monitoring Rail Infrastructure: Application of Composite Indicators. International Journal of Rail Transportation, Vol. 3, 2015, pp. 61–80.

12.

Chellaswamy

Akila

Babu

A. D.

Arasan

N. K.

Fuzzy Logic Based Railway Track Condition Monitoring System. Proc., IEEE International Conference on Communication and Nanotechnology (ICE-CCN), Tirunelveli, India, IEEE, New York, 2013. https://doi.org/10.1109/ICE-CCN.2013.6528503.

13.

BRSSOS Ltd. Track Monitoring. BRSSOS Ltd., 2015. http://www.brssos.co.uk/track-monitoring.html. Accessed May 2018.

14.

Lebel

Soize

Funfschilling

Perrin

High-Speed Train Suspension Health Monitoring Using Computational Dynamics and Acceleration Measurements. Vehicle System Dynamics, Vol. 58, No. 6, 2019, pp. 911–932. https://doi.org/10.1080/00423114.2019.1601744.

15.

Wang

Cheng

Allen

Yin

Zou

Zhang

Analysis of Vibration and Temperature on the Axle Box Bearing of a High-Speed Train. Vehicle System Dynamics, Vol. 58, No. 10, 2019, pp. 1605–1628. https://doi.org/10.1080/00423114.2019.1645340.

16.

Chellaswamy

Krishnasamy

Balaji

Dhanalakshmi

Ramesh

Optimized Railway Track Health Monitoring System Based on Dynamic Differential Evolution Algorithm. Measurement, Vol. 152, 2020, p. 107332. https://doi.org/10.1016/j.measurement.2019.107332.

17.

Allotta

D’Adamio

Marini

Meli

Pugi

Rindi

A New Strategy for Dynamic Weighing in Motion of Railway Vehicles. IEEE Transactions on Intelligent Transportation Systems, Vol. 16, 2015, pp. 3520–3533. https://doi.org/10.1109/TITS.2015.2477104.

18.

Shafahi

Hakhamaneshi

Application of a Maintenance Management Model for Iranian Railways Based on the Markov Chain and Probabilistic Dynamic Programming. International Journal of Scientific & Technology, Transaction A: Civil Engineering, Vol. 16, No. 1, 2009, pp. 87–97.

19.

Podofillini

Zio

Vatn

Risk-Informed Optimisation of Railway Tracks Inspection and Maintenance Procedures. Reliability Engineering & System Safety, Vol. 91, No. 1, 2006, pp. 20–35. https://doi.org/10.1016/j.ress.2004.11.009.

20.

Prescott

Andrews

Modelling Maintenance in Railway Infrastructure Management. Proc., 2013 Annual Reliability and Maintainability Symposium (RAMS), Orlando, FL, IEEE, New York, 2013. https://doi.org/10.1109/RAMS.2013.6517678.

21.

Parhi

D. R.

Dash

A. K.

Das

H. C.

Formulation of a Genetic Algorithm Based Methodology for Multiple Crack Detection in a Beam Structure. Australian Journal of Structural Engineering, Vol. 12, No. 2, 2015, pp. 127–139.

22.

Xue

Tang

Zhou

Identification of Structural Systems Using Particle Swarm Optimization. Journal of Asian Architecture and Building Engineering, Vol. 8, No. 2, 2018, pp. 517–524.

23.

Xiukun

Feng

Limin

Urban Rail Track Condition Monitoring Based on In-Service Vehicle Acceleration Measurements. Measurement, Vol. 80, 2016, pp. 217–228.

24.

Chen

You

Ling

Liang

Zimmermann

Dynamic Urban Surveillance Video Stream Processing Using Fog Computing. Proc., IEEE 2nd International Conference on Multimedia Big Data, Taipei, IEEE, New York, 2016, pp. 105–112. https://doi.org/10.1109/BigMM.2016.53.

25.

Jian

Xin

Xue

You

D. Z.

Cloud Computing and Visual Attention Based Object Detection for Power Substation Surveillance Robots. Proc., IEEE 28th Canadian Conference on Electrical and Computer Engineering (CCECE), Halifax, NS, IEEE, New York, 2015. https://doi.org/10.1109/CCECE.2015.7129299.

26.

Sun

Zhang

Bie

Advancing Researches on IoT Systems and Intelligent Applications. Personal and Ubiquitous Computing, Vol. 22, 2018, pp. 449–452. https://doi.org/10.1007/s00779-018-1159-9.

27.

Zuo

Tao

Nee

A. Y. C.

An Internet of Things and Cloud-Based Approach for Energy Consumption Evaluation and Analysis for a Product. International Journal of Computer Integrated Manufacturing, Vol. 31, No. 4–5, 2017, pp. 337–348.

28.

Trappey

A. J. C.

Trappey

C. V.

Fan

C.-Y.

Hsu

A. P. T.

X.-K.

Lee

I. J. Y.

IOT Patent Roadmap for Smart Logistic Service Provision in the Context of Industry 4.0. Journal of the Chinese Institute of Engineers, Vol. 40, No. 7, 2017, pp. 593–602.

29.

Huang

H.-P.

Yan

J.-L.

Haung

T.-H.

Haung

M.-B.

IoT-Based Networking for Humanoid Robots. Journal of the Chinese Institute of Engineers, Vol. 40, No. 7, 2017, pp. 603–613.

30.

Wikipedia. Lists of Rail Accidents. 2016. https://en.wikipedia.org/wiki/List_of_rail_accidents(2010-present). Accessed April 2018.

31.

Gao

Wang

Yao

Self-Powered ZigBee Wireless Sensor Nodes for Railway Condition Monitoring. IEEE Transactions on Intelligent Transportation Systems, Vol. 99, 2017, pp. 1–10.

32.

Arun

K. S.

Swarup

Agarwal

Singh

Vision Based Rail Track Extraction and Monitoring Through Drone Imagery. ICT Express. Vol. 5, No. 4, 2019, pp. 250-255. https://doi.org/10.1016/j.icte.2017.11.010

33.

Mohammad

A. S.

Mehdi

Intelligent Condition Monitoring of Railway Signaling in Train Detection Subsystems. Journal of Intelligent & Fuzzy Systems. Vol. 24, No. 4, 2013, pp. 859-869. https://doi.org/10.3233/IFS-2012-0604

34.

Mahajan

Zong

Energy Efficiency Analysis of Query Optimizations on MongoDB and Cassandra. Proc., 8th International Green and Sustainable Computing Conference, Orlando, FL, IEEE, New York, 2017, pp. 1–6. https://doi.org/10.1109/IGCC.2017.8323581.

35.

Anand

Rao

C. M.

MongoDB and Oracle NoSQL: A Technical Critique for Design Decisions. Proc., International Conference on Emerging Trends in Engineering, Technology and Science, Pudukkottai, India, IEEE, New York, 2016, pp. 1–4. https://doi.org/10.1109/ICETETS.2016.7602984.

36.

Chellaswamy

Rao, Taha

Y. R.

Jawwad

Shaji

Sharma

A Novel Optimization Method for Parameter Extraction of Industrial Solar Cells. Proc., IEEE International Conference on Innovations in Power and Advanced Computing Technologies (i-PACT), Vellore, India, IEEE, New York, 2019, pp. 1–6. https://doi.org/10.1109/i-PACT44901.2019.8959966.

37.

Rohani

Gingras

Gruyer

A Novel Approach for Improved Vehicular Positioning Using Cooperative Map Matching and Dynamic Base Station DGPS Concept. IEEE Transactions on Intelligent Transportation Systems, Vol. 17, No. 1, 2016, pp. 230–239.

38.

Chellaswamy

Srinivasan

Babu

R. R.

A Humanitarian Intelligent Level Crossing Controller Utilizing GPS and GSM. Proc., 2014 IEEE Global Humanitarian Technology Conference - South Asia Satellite (GHTC-SAS), Trivandrum, India, IEEE, New York, 2014, pp. 245–250. https://doi.org/10.1109/GHTC-SAS.2014.6967591.