According to the Federal Railroad Administration (FRA) Highway-Rail Grade Crossing Accident/Incident database, more than 12,000 accidents occurred between 2012 and 2017 in the United States with casualties of around 3900. Despite repeated efforts to fully understand the risk factors that contribute to highway-rail grade crossing collisions, there still remain many uncertainties. A machine learning approach is proposed in this paper to find out significant factors, along with their individual impacts of crash severities at grade crossings. One of the most efficient and accurate machine learning algorithms, extreme gradient boosting (XGB or XGBoost), is applied to analyze 21 different accident and crossing -related characteristics per driver severities. The XGB model has been proven in previous studies across many research areas in transportation to outperform other machine learning-based methods and statistical classification methods, such as multinomial logit model, multiple additive regression trees, decision tree, and random forest, especially in prediction accuracy. Thereby, applying the algorithm is expected to provide highly reliable results to identify important factors that have impacts on injury severities at grade crossings. Such application will further aid the discovery of potential crossings with significant factors. The FRA’s Highway-Rail Grade Crossing Accident/Incident database from 2012 to 2017 is fused with the FRA Highway-Rail Crossing Inventory database for the analysis. Observations with missing information were removed from the original database. Crossing position under or over the railroad and pedestrian or other types of highway users were also not considered since they were not specifically of interest in this study. After the database cleaning process, it condensed to the total of 1,250 accidents out of the retrieved 12,630 from the combined database. The results show that adjacent highway traffic volume and train speed are the most significant factors causing accidents and injury severity. They are followed by the driver’s age and the estimated vehicle speed. It also indicated that truck-involved accidents and crossings with gates, flashing lights, and other types of warning devices combined, and highway user’s gender as a male also pertain to the higher injury rate. Through this study, it is possible to provide guidance to decision-makers in recognizing possible risks at-grade crossings that may cause driver casualties.

Rolling stock, particularly of freight railroads, is currently maintained using regular preventative and corrective maintenance schedules. This maintenance approach recommends sets of inspections and maintenance procedures based on the average expected wear and tear across their inventory. In practice, however, this approach to scheduling preventative maintenance is not always effective. When scheduled too soon, it results in a loss of operating revenue, whereas when it is scheduled too late, equipment failure could lead to costly and disastrous derailments. Instead, proactive maintenance scheduling based on Big Data Analytics (BDA) could be utilized to replace traditional scheduling, resulting in optimized maintenance cycles for higher train safety, availability, and reliability. BDA could also be used to discover patterns and relationships that lead to train failures, identify manufacturer reliability concerns, and help validate the effectiveness of operational improvements. In this work, we introduce a train inventory simulation platform that enables the modelling of different train components such as wheels, brakes, axles, and bearings. The simulator accounts for the wear and tear in each component and generates a comprehensive data set suitable for BDA that can be used to evaluate the effectiveness of different BDA approaches in discerning patterns and extracting knowledge from the data. It provides the basis for showing that BDA algorithms such as Random Forest [9] and Linear Regression can be utilized to create models for proactive train maintenance scheduling. We also show the capability of BDA to detect hidden patterns and to predict failure of train components with high accuracy.

Railway track vertical alignment is an important indicator of serviceability condition and thus plays a critical role for maintenance planning. Estimating the rail profile through the vertical acceleration readings provides an efficient alternative to the current practice of optical methods using special vehicles. This paper proposes an algorithm to estimate the rail vertical profile using the vertical acceleration of the vehicle resulting from the train-track dynamic interaction. The algorithm is designed to approximate the inverse of the transfer function from the rail vertical roughness to the train’s measured acceleration. The approximation problem is formulated into an H-infinity optimal control design problem, which can be further transferred into a problem of convex optimization. The proposed algorithm possesses several advantages including easy design, little tuning effort, and low computational cost. In addition, to take into account the model uncertainty, an optimization-based learning framework is proposed to further enhance the performance of the proposed algorithm. The numerical study has been conducted comprehensively to validate the observer’s properties and effectiveness in reconstructing of the rail vertical roughness.

Designed for detecting train presence on tracks, track circuits must maintain a level of high availability for railway signaling systems. Due to the fail-safe nature of these critical devices, any failures will result in a declaration of occupancy in a section of track which restricts train movements. It is possible to automatically diagnose and, in some cases, predict the failures of track circuits by performing analytics on the track signals. In order to perform these analytics, we need to study the coded signals transmitted to and received from the track. However, these signals consist of heterogeneous pulses that are noisy for data analysis. Thus, we need techniques which will automatically group homogeneous pulses into similar groups. In this paper, we present data cleansing techniques which will cluster pulses based on digital analysis and machine learning. We report the results of our evaluation of clustering algorithms that improve the quality of analytic data. The data were captured under revenue service conditions operated by Alstom. For clustering algorithm, we used the k -means algorithm to cluster heterogeneous pulses. By tailoring the parameters for this algorithm, we can control the pulses of the cluster, allowing for further analysis of the track circuit signals in order to gain insight regarding its performance.

1 In this paper we present an optimization based technique to allocate time windows for railroad track maintenance in order to minimize the impact on train circulation. The hypothesis is that it is possible to reduce the impact on train circulation if track maintenance slots are scheduled in naturally occurring gaps on the circulation, instead of just using a decision making process that does not take the train circulation into consideration. The proposed solution uses simulated annealing as the basis for the algorithm and yields promising results when a random initial placement of track blocks is used. We show the results of our experiments that consider as restrictions the earliest initial time for the track block and the latest termination time for the track blocks (to account for restriction in crew work hours, for example). For this work we do not restrict the number of simultaneous maintenance teams that could be working, future investigations will account for that restriction.

Train accidents can be attributed to human factors, equipment factors, track factors, signaling factors, and Miscellaneous factors. Not only have these accidents caused damages to railroad infrastructure and train equipment leading to excessive maintenance and repair costs, but some of these have also resulted in injuries and loss of lives. Big Data Analytics techniques can be utilized to provide insights into possible accident causes, thus resulting in improving railroad safety and reducing overall maintenance expenses as well as spotting trends and areas of operational improvements. We propose a comprehensive Big Data approach that provides novel insights into the causes of train accidents and find patterns that led to their occurrence. The approach utilizes a combination of Big Data algorithms to analyze a wide variety of data sources available to the railroads, and is being demonstrated using the FRA train accidents/incidents database to identify factors that highly contribute to accidents occurring over the past years. The most important contributing factors are then analyzed by means of association mining analysis to find relationships between the cause of accidents and other input variables. Applying our analysis approach to FRA accident report datasets we found that railroad accidents are correlating strongly with the track type, train type, and train area of operation. We utilize the proposed approach to identify patterns that would lead to occurrence of train accidents. The results obtained using the proposed algorithm are compatible with the ones obtained from manual descriptive analysis techniques.

A micromechanical-based 2D framework is presented to study the rolling contact fatigue (RCF) in rail steels using finite element method. In this framework, the contact patch of rail and wheel is studied by explicitly modeling the grains and grain boundaries, to investigate the potential origin of RCF at the microstructural level. The framework incorporates Voronoi tessellation algorithm to create the microstructure geometry of rail material, and uses cohesive zone approach to simulate the behavior of grain boundaries. To study the fatigue damage caused by cyclic moving of wheels on rail, Abaqus subroutines are employed to degrade the material by increasing the number of cycles, and Jiang-Sehitoglu fatigue damage law is employed as evolution law. By applying Hertzian moving cyclic load, instead of wheel load, the effect of traction ratio and temperature change on RCF initiation and growth are studied. By considering different traction ratios (0.0 to 0.5), it is shown that increasing traction ratio significantly increases the fatigue damage. Also by increasing traction ratio, crack initiation migrates from the rail subsurface to surface. The results also show that there are no significant changes in the growth of RCF at higher temperatures, but at lower temperatures there is a measurable increase in RCF growth. This finding correlates with anecdotal information available in the rail industry about the seasonality of RCF, in which some railroads report noticing more RCF damage during the colder months.

A ballast layer is used to facilitate drainage and load transferring in railroad track structure. With tonnage accumulation, fine materials, such as coal dust, clay, locomotive sand, degraded ballast aggregate, and other small particles, will penetrate into the clean and uniformly graded ballast layer causing contamination, usually referred as fouling. Fouling is unfavorable to railroad track performance due to the reduced drainage and consequent engineering challenges including but not limited to mud pumping, excessive settlement, and reduced bearing capacity. Previous research has investigated the mechanical behavior of the fouled ballast in both the laboratory and the field environment. However, the fundamental mechanism that governs the manner in which the fouling materials are transported and accumulated in the ballast layer is not thoroughly understood. Researchers at the University of South Carolina have initiated the effort to investigate the fouling process in the ballast layer. High-fidelity computational fluid dynamics (CFD) simulation is used to study the fluid flow patterns in order to quantify the transport behavior of the fine particles within the ballast layer and potential impact to the track performance and drainage. Specifically, the ballast layer is treated as a porous material, and the fouling materials are modeled as distinct individual particles to assess the probability of their trajectory location. This paper presents the preliminary results of the simulated path of the fouling materials in the ballast layer under seepage, and demonstrates the capability of the developed algorithm to quantify the effects of the ballast layer characteristics on fouling materials transport. The findings from this study will be beneficial for optimizing shoulder ballast cleaning or undercutting practices.

In a typical railyard environment, a myriad of large and dynamic objects pose significant risks to railyard workers. Unintentional falls, trips and collisions with dynamic rolling stock due to distractions or lack of situational awareness are an unfortunate reality in modern railyards. The challenges of current technologies in detecting and tracking multiple differently-sized mobile objects in situations such as i) one-on-one, ii) many-to-one, iii) one-to-many, iv) blind spot, and v) interfering/non-interfering separation creates the possibility for reduction or loss of situational awareness in this fast-paced environment. The simultaneous tracking of assets with different size, velocity and material composition in different working and environmental conditions can only be accomplished through joint infrastructure-based asset discovery and localization sensors that cause no interference or impediment to the railyard workers, and which are capable of detecting near-misses as well. Our team is investigating the design and performance of such a solution, and is currently focusing on the innovative usage of lightweight low-cost RADAR under different conditions that are expected to be encountered in railyards across North America. We are employing Ancorteks 580-AD Software Defined RADAR (SDRadar) system, which operates at the license-free frequency of 5.8 GHz and with a variety of different configuration options that make it well-suited for generalized object tracking. The challenges, however, stem from the unique interplay between tracking large metallic objects such as railcars, locomotives, and trucks, as well as smaller objects such as railyard workers, in particular their robust discernment from each other. Our design’s higher-level system can interact with the lower-level SDRadar design to change the parameters in real-time to detect and track large objects over significant distances. The algorithm optimally adjusts waveform, sweep time and sample rate based on one or multiple detected object cross-sections and subsequently alters these parameters to be able to discern other objects from them that are in close proximity. We also use an ensemble method to determine the velocity and distance of target objects to accurately track the subject and larger objects at a distance. The methodology has been field-tested with several test cases in a multitude of weather and lighting conditions. We have also tested the proper height, azimuth and elevation angles for positioning our SDRadar to alleviate the risk of blind spots and enhancing the detection and tracking capabilities of our algorithm. The approach has outperformed our previous tests using visual and acoustic sensors for detection and tracking railroad workers in terms of accuracy and operating flexibility. In this paper, we discuss the details of our proposed approach and present our results from the field tests.

Early detection of rail defects can avoid derailments and costly damage to the train and railway infrastructure. Small breaks, cracks or corrugations on the rail can quickly propagate after only a few train cars have passed over it, creating a potential derailment. The current technology makes use of a dedicated instrumented car or a separate railway monitoring vehicle to detect large breaks. These cars are usually equipped with accelerometers mounted on the axle or side frame. The simple detection algorithms use acceleration thresholds which are set at high values to eliminate false positives. As a result, rail surface defects that produce low amplitude acceleration signatures may not be detected, and special track components that produce high amplitude acceleration signatures may be flagged as defects. This paper presents the results of a feasibility study conducted to develop new and more advanced sensory systems as well as signal processing algorithms capable of detecting various rail surface irregularities. A dynamic wheel-rail interaction model was used to simulate train dynamics as a result of rail defects and to assess the potential of this new technology on rail defect detection. In a future paper, we will present experimental data in support of the proposed model and simulations.

Cameras are substantial elements to provide security to passengers in public spaces, e.g. in urban subway environments. Automated image processing algorithms are used more and more often to analyze the cameras’ video streams. However, misaligned cameras may produce serious problems by either generating false alarms or even being blind due to a shifted field of view. This paper presents two simple, real-time, and straight-forward software methods to detect camera movement and also allow for an adjustable tolerance, i.e. small changes are acceptable. The first approach shifts a reference edge (from a master image) along a convenient path; the second method uses distance measuring to detect a critical camera movement. Both methods are tested on real-world video from a subway environment. Since the algorithms are proposed for application in a railway environment with typically high requirements on operational robustness and reliability, special emphasis is put on constraints and limitations of the presented algorithms.

The variations in the operation timetable or schedule of an electrified transit rail system can lead to substantial fluctuations in power demands of its traction power network. This paper studies the correlation between the maximum power demands and timetable perturbations for electrified transit rail systems. Specifically, the operation schedule uncertainties are quantified as two parameters: headway shift and headway perturbation. A computation algorithm is introduced to illustrate how to use these two parameters to obtain the worst case scenario to obtain maximum power demand of traction power substations. Also a special type of catenary-free light rail system is used as an example to illustrate the algorithm and numerical results.

First-mile and last-mile travel is the bottleneck of using the public transportation service. This paper considers the passenger matching and vehicle routing problem in the first-mile ridesharing service connecting train schedules. Then a mixed integer model is proposed to formulate the problem. Since the problem is NP hard, we develop a simulated annealing (SA) algorithm with four neighborhood structures to solve this problem. Experiments are designed to test the proposed model and algorithm. The experimental results verify the effectiveness of the algorithm and demonstrate that the proposed algorithm can obtain satisfactory solutions within a reasonably short time.

Two major components of rolling stock that are always of great interest when it comes to maintenance and safety related issues are car wheels and bearings. Rail car wheels are subjected to a variety of damage types due to their interaction with the track and brakes. It is important for the rail industry to detect these defects and take proper action at an early stage, before more damage can be caused to the train or possibly the track and to prevent possible safety hazards. Different inspection sensors and systems, such as wheel impact monitors, wheel profile detectors, hotbox detectors and acoustic detection technologies, are employed to detect different types of wheel and bearing defects. Usually no single sensor can accurately detect all kinds of damages and hence a combination of different sensors and systems and manual inspection by experts is used for wheel maintenance purposes and to guarantee train safety. The more complete and accurate the automatic defect detections are, the less manual examination is necessary, leading to potential savings in inspection time/resources and rail car maintenance costs. Wayside thermal and visible spectrum cameras are one option for the automatic wheel and bearing inspection. Each of these sensors has their own strengths and weaknesses. There are some types of defects that are not detectable at an early stage in the images taken by a vision camera, however these defects generate a distinctive heat pattern on the wheel or bearing that is clearly visible in the thermal imagery. On the other hand, other damages might be detectable from the visible spectrum image, but not necessarily have a distinguishable heat pattern in the thermal imagery. Since a thermal image is basically built of solely temperature data, it excludes other critical information, such as texture or color. This makes thermal and visible spectrum imagery complementary and if the images are fused the result will benefit from the strengths of both sensors. In this paper, wavelet decomposition is employed to extract the features of the thermal and vision imagery. Then the two images are merged based on their decompositions and a fused image is composed. The resulting fused image contains more information than each individual image and can be used as an input for image-based wheel and bearing defect detection algorithms. To verify the proposed method and to show an example of this application, it is demonstrated on a real data set from a Union Pacific rail line to identify sliding wheels.

The University of California at San Diego (UCSD), under a Federal Railroad Administration (FRA) Office of Research and Development (R&D) grant, is developing a system for high-speed and non-contact rail defect detection. A prototype using an ultrasonic air-coupled guided wave signal generation and air-coupled signal detection, paired with a real-time statistical analysis algorithm, has been realized. This system requires a specialized filtering approach based on electrical impedance matching due to the inherently poor signal-to-noise ratio of air-coupled ultrasonic measurements in rail steel. Various aspects of the prototype have been designed with the aid of numerical analyses. In particular, simulations of ultrasonic guided wave propagation in rails have been performed using a Local Interaction Simulation Approach (LISA) algorithm. The system’s operating parameters were selected based on Receiver Operating Characteristic (ROC) curves, which provide a quantitative manner to evaluate different detection performances based on the trade-off between detection rate and false positive rate. The prototype based on this technology was tested in October 2014 at the Transportation Technology Center (TTC) in Pueblo, Colorado, and again in November 2015 after incorporating changes based on lessons learned.

Due to the increasing scarcity of fossil fuels and the climate change, the importance of energy efficiency is increasing. This importance is major especially in areas where the energy consumption is high. Rail transport depicts such an area. The highest proportion of energy consumed in the railway is the so called traction energy. This energy is required for the train run. In the timetable, allowances leave a margin for the driving style of train run. By the selective use of strategies that change the driving style, it is possible to exploit these allowances and reduce the traction energy consumption. The first objective of this study deals with the development of algorithms for energy-saving driving style. First, the necessary input variables of the algorithms based on the literature research and the formulas of train dynamics were determined. Then the algorithms were developed to create different energy-saving driving styles, resulting choose the best result which should be shown as a driving recommendation. The developed algorithms were used in an application example in order to calculate the potential of energy-savings. The example should represent the influence of the input variables for a comparison of different situations. At last the acceptance of the determined driving strategies in practice was investigated. By implementing the design thinking method it was identified that driver advisory systems and training programs are necessary to facilitate energy-saving driving in practice.

Train car wheels are subjected to different types of damages due to their interactions with the brake shoes and track. If not detected early, these defects can worsen, possibly causing damage to the bogie and rail. In the worst-case scenario, this rail damage can possibly lead to later derailments, a serious concern for the rail industry. Therefore, automatic inspection and detection of wheel defects are high priority research areas. An automatic detection system not only can prevent train and rail damage, but also can reduce operating costs as an alternative for tedious and expensive manned inspection. The main contribution of this paper is to develop a computer vision method for automatically detecting the defects of rail car wheels using a wayside thermal camera. We concentrate on identification of flat-spotted/sliding wheels, which is an important issue for both wheel and suspension hardware and also rail and track structure. Flat spots occur when a wheel locks up and slides while the vehicle is still moving. As a consequence, this process heats up local areas on the metal wheel, which can be observed and potentially detected in thermal imagery. Excessive heat buildup at the flat spot will eventually lead to additional wheel and possibly rail damage, reducing the life of other train wheels and suspension components, such as bearings. Furthermore, as a byproduct of our algorithm, we propose a method for detecting hot bearings. A major part of our proposed hot bearing detection algorithm is common with our sliding wheel detection algorithm. In this paper, we first propose an automatic detection and segmentation method that identifies the wheel and bearing portion of the image. We then develop a computer vision method, using Histogram of Oriented Gradients to extract features of these regions. These feature descriptors are input to a Support Vector Machine classifier, a fast classifier with a good detection rate, which can detect abnormalities in the wheel. We demonstrate our methods on several real data sets taken on a Union Pacific rail line, identifying sliding wheels and hot bearings in these images.

In current rail inspection processes, following a detection of a suspected internal defect, an additional secondary detailed inspection is required to (1) confirm the presence of the flaw and (2) determine the severity of the flaw to allow for optimal post-detection rail maintenance planning. Current ultrasonic devices in this secondary inspection efforts heavily rely the expertise and experience of the test personnel’s judgement to confirm the rail flaw and to characterize the internal defect by analyzing reflected waveforms. To eliminate the uncertainties in this secondary inspection process and to provide the testing operators with better defect characterization such as the size and location of the flaw, a defect ultrasonic imaging device utilizing synthetic aperture focusing (SAF) techniques is proposed in this paper. These imaging techniques have been successfully demonstrated in medical imaging, providing quantitative characterization of internal components, allowing for a better prognosis. Ultimately, having a quantitative evaluation of the internal flaw can lead to an increase in the safety of train operations by preventing derailments. Thus, in this paper, a preliminary portable rail defect imaging concept is proposed by the University of California, San Diego, to provide three dimensional images of internal defects in the rail. The prototype reconstructs a three-dimensional volumetric image of the rail, utilizing multiple two-dimensional planar ultrasonic images. Improvements to the conventional tomographic imaging algorithms have been made by utilizing a mode-selective image reconstruction scheme that exploits the specific displacement field, respectively, of the longitudinal wave modes and the shear wave modes, both propagating simultaneously in the test volume. The specific mode structure is exploited by an adaptive weight assignment to the ultrasonic tomographic array. Such adaptive weighting forces the imaging array to look at a specific scan direction and better focus the imaging onto the actual flaw (ultrasound reflector). This preliminary study shows that the usages of the adaptive weights based on wave structure improves image dynamic range and spatial resolution, when compared to a conventional ultrasonic imaging technique such as Delay-And-Sum (DAS). Results will be shown both from numerical models and experimental tests of internal flaws in rails.

Short pitch corrugation is commonly seen in all kinds of tracks. There is not yet a conclusive explanation in the literature for its initiation and growth mechanisms. In this paper, we use an axle box acceleration (ABA) measurement system to detect corrugation. ABA can be easily implemented in operational trains, providing direct and reliable health monitoring of the track. We have extended a detection algorithm for rail surface local short wavelength defects to also detect short pitch corrugation, which is a continuous defect over the track. A 3D transient FE wheel-track model is employed to find theoretical signature tunes of the wheel-track system response when passing over a short pitch corrugation. Numerical simulations agree with ABA measurement obtained in the Dutch rail network. Based on the signature tune identified, an automatic detection algorithm is developed. Preliminary results with the algorithm are discussed. Field observations show a good potential of the detection algorithm to be used by inframanagers, to detect and monitor corrugation.

Hardware-in-the-Loop (HIL) testing allows for scaled to full-scale evaluation of components without constructing full-scale vehicle prototypes. HIL is also a rapid means of performance evaluation of difficult to model components, in real-time simulation with up to full-scale components providing their own response in place of a model. HIL also provides a working tool for control algorithm refinements prior to full system integration. In hybrid and electric locomotive applications this is particularly important while choosing the battery chemistry and sizing the energy storage system. Energy storage HIL testing permits optimal sizing of battery power and energy density along with management algorithm tuning. In general, HIL provides insight into optimal locomotive powertrain design. After the development of full EV and HEV locomotive models, battery component models were replaced with real-time interfaces to lead carbon and lithium cobalt batteries, which validated the battery models and helped tune vehicle controllers for direct application onto the vehicle platform.