The present research seeks to compare the flexural behavior of different crosstie technologies with a goal of understanding the similarities and differences in flexural behavior. Commercially produced wood, composite, and concrete crossties are tested in center negative and rail-seat positive flexure as described in AREMA Chapter 40 and the relative flexural performances are compared. Unsurprisingly, wood and composite ties have lower bending stiffness and somewhat lower strength while demonstrating greater deflection capability. The observations obtained in this testing help to quantify the relative structural contribution of the differing tie technologies and may indicate that required flexural capacity for concrete ties should be assessed on the basis of deflection rather than load. Accordingly, we suggest further research into concrete tie design paradigms that prioritize deflection performance which we believe will lead to robust performance and long service life.
Category: Project Abstracts
Explainable Spatiotemporal Machine Learning for Multicomponent Railway Track Degradation Modeling
Modern railway inspection systems generate large volumes of high-resolution data describing the condition of individual track components, including rails, crossties, fasteners, ballast, joints, and track geometry. While these data provide detailed information about infrastructure condition, they are often analyzed using component-specific thresholds or independent defect assessments, limiting our understanding of how deficiencies evolve over time and interact across neighboring track sections.
This project proposes an explainable spatiotemporal machine learning framework to analyze railway inspection data and uncover temporal and spatial patterns of track deterioration. The research addresses two fundamental questions: (1) Can historical measurements of individual track deficiencies be used to predict their future condition? (2) Do spatial relationships and interactions among multiple deficiencies improve our understanding and prediction of deterioration compared
with analyzing defects independently?
The study will utilize a large-scale inspection dataset comprising consecutive track sections with repeated measurements of track geometry, rail wear, crosstie condition, ballast characteristics, fasteners, anchors, and other infrastructure components. Segment-level features will characterize defect severity, spatial continuity, local variability, and interactions among neighboring components. Statistical methods and interpretable machine learning models will be employed to identify significant spatiotemporal relationships and forecast future deterioration of measured track conditions. Explainable AI techniques will quantify the influence of individual variables and their interactions, while spatially blocked validation will ensure reliable model evaluation.
The proposed research will advance understanding of railway deterioration by treating track defects as an interconnected spatiotemporal process rather than isolated observations. Expected outcomes include predictive models for the evolution of individual deficiencies, quantitative characterization of spatial and temporal degradation patterns, and improved insight into relationships among track components. The resulting framework will provide transportation agencies with a scientifically grounded approach for extracting actionable information from large-scale inspection datasets, supporting more proactive inspection planning, maintenance scheduling, and long-term infrastructure management.
Temporal Fusion Transformer-Based Grade Crossing Signal Timing Prediction Using Train Trajectory Data
Direct vehicle warnings offer great potential toward improved safety by preventing or warning highway vehicles entering crossings at the time of train arrivals. However, this requires high ontime locational accuracy and timely communications. This project aims to work towards improved accuracy by developing an advanced, uncertainty-aware prediction model for grade crossing Signal Phase and Timing (SPaT) using train trajectory prediction. Accurate prediction of train arrival and departure times at highway-rail grade crossings remains challenging due to operational uncertainties such as varying speeds, dwell times, train length, and network interactions. Existing systems rely on Kalman filtering using real-time train location data from the head-of-train (HOT) and end-of-train (EOT), but these approaches have limited capability in
capturing nonlinear dynamics and uncertainty.
Building on prior work, the research team has already developed a Kalman-filter-based baseline model and collected train trajectory datasets from the Transportation Technology Center (TTC), Pueblo, CO. This project will leverage these datasets along with simulation data to develop and evaluate a hybrid probabilistic prediction framework.
The proposed framework integrates track-constrained state-space modeling with a Temporal Fusion Transformer (TFT) for multi-horizon forecasting. To explicitly address uncertainty, the model incorporates quantile prediction and conformal calibration to generate probabilistic predictions with calibrated confidence levels.
Rather than producing a single deterministic estimate, the system will generate multiple likely train trajectories and associated confidence levels. These predictions are used to derive SPaT at grade crossings. All evaluations will be conducted in a simulation environment using TTC datasets to assess prediction accuracy, robustness, and uncertainty calibration.
By leveraging existing datasets and focusing on simulation-based validation, this project reduces development risk while advancing beyond current methods. The outcomes will improve SPaT reliability, enhance situational awareness, and reduce safety risks and traffic congestion at highway-rail grade crossings.
Detectability-driven Design of An Easy-to-Deploy Wayside Hunting Detection System
Hunting (self-excited oscillation of rolling stock) is a typical example of harmful vehicle-track interaction. Hunting gives rise to wheel-rail dynamic forces, increasing derailment and buckling risks and accelerating wear and degradation. Wayside hunting detection can identify railcars exhibiting poor stability, thereby improving safety and supporting maintenance decision-making. However, existing systems, including truck hunting detectors (THD), truck performance detectors (TPD), and wheel impact load detectors (WILD), rely on track-mounted instrumentation, such as strain gauges or near-rail optical sensors. These systems require installation, calibration, and
maintenance within the railroad right-of-way, limiting their deployability and relocatability.
This project explores an easy-to-deploy wayside hunting detection concept based on Laser Doppler Vibrometry (LDV). LDV enables non-contact measurement of dynamic responses at standoff distances of several meters or more, and its use for wayside inspection of rolling stock remains underexplored. This remote sensing capability eliminates the need for on-track instrumentation, enabling flexible deployment, rapid relocation, and expanded coverage. Wayside LDV can directly capture the dynamic response of passing railcars, offering a new and potentially more accurate pathway for assessing rolling stock stability.
Despite these advantages, several technical challenges must be addressed in the design and development of such an LDV-based hunting detection system, including:
1) Variability of hunting behaviors: hunting behavior varies significantly across railcar types, speeds, wheel profiles, and loading conditions, resulting in diverse dynamic signatures. Moreover, vehicle-track interaction introduces additional stochastic vibration behaviors.
2) Laser speckle noise: high-speed scanning of a coherent laser on railcar surfaces leads to drastic changes in laser interference patterns, generating so-called speckle noise. Speckle noise elevates the noise floor of LDV measurements and complicates feature extraction.
3) Uncertainties in detectability: high speed usually triggers hunting, but at the same time, shortens the observation windows per railcar, reducing the frequency analysis resolution. This, combined with stochastic vehicle vibrations and measurement noise, poses challenges to reliable hunting detection.
This project develops a detectability-driven framework that integrates vehicle dynamics with LDV measurement uncertainty to study the detectability of the proposed LDV-based wayside hunting detection concept.
Framework and Case Studies for Improved Estimating and Modeling Intercity Train Passenger Ridership
Demand for Intercity Passenger Rail (IPR) transportation in the US continues to grow as the National Railroad Passenger Corporation (Amtrak) sets record annual ridership figures year after year. Since forecasted ridership is critical to developing and evaluating proposals for new or expanded IPR service across the country, an improved understanding of what factors drive origin destination (O-D) passenger travel demand for IPR service is essential for planners, stakeholders, and operators. The O-D of intercity passenger train riders can be obtained from ticket sales, but comprehensive O-D US IPR ridership data, or the trip purposes of riders – which drives their
travel decision making – are not. Therefore, additional analysis and modeling is required to make the leap from station-level demand to ridership along each segment of an IPR corridor and to tailor and market intercity rail services to maximize ridership for any given level of investment.
This research addresses this gap through two methodological approaches. Nationally, it leverages various metrics for station-level ridership, corridor passenger-miles, average trip lengths, and the relative ridership rankings of O-D station pairs that are publicly available for various US IPR corridors and services. The work will be complemented by a detailed collection of ridership data on the three intercity passenger services offered in the State of Michigan, first as an online survey accessed via mobile devices to be completed by passengers during active train rides, and then through in-depth analysis of in-person interviews by Michigan Tech researchers with passengers at rail stations and aboard the train. A geospatial analysis will be conducted with the survey results to identify the origins, destinations, and travel mode choices for riders to and from rail stations.
The results from this research can provide stakeholders involved in the development of passenger rail corridors a better understanding of ridership factors compared to the limitations of the current station-only approach. The comparison of modeling framework with actual case study outcomes helps to build confidence in the model and the resulting model will help inform trade-offs between making investments to add frequencies or infill stations to existing services or developing new routes and corridors.
Modeling the Reliability of Shared Corridors Under Contemporary Operating Practices
The US rail network includes shared corridors that feature both freight and passenger train operations. The reliability of these mainline links is critical to ensuring that freight supply chains meet the level of service demanded by industrial customers while also providing performance that meets the mobility needs of commuters and the traveling public. Satisfying both of these rail transport objectives promotes economic development and supports job creation. However, the differing characteristics and performance requirements of freight and passenger trains create numerous operational challenges that impact the reliability of both operations. Optimizing the
level of capital investment in track infrastructure and train control system required to meet the desired reliability of a shared corridor is critical to the economics of these operations.
Previous researchers have used simulation to document the capacity (throughput) and performance (delay) implications of differences in passenger and freight train speed, power, weight, length and priority. Most critically, the capacity consumed by a single train varies across train types, and corridors with increasing levels of train heterogeneity require correspondingly more infrastructure to achieve the same performance and reliability as heterogenous freight-only or passenger-only corridors. However, a key limitation of existing research is that, during the 10 to 15 years since it has been conducted, freight railroads have altered their operations considerably, shifting to widespread deployment of longer trains, relying on locomotive assignments with lower horsepower-per-ton, and reconstructing corridors with longer passing
sidings or stretches of two main tracks in place of single track with frequent short sidings. In addition, the relative priority or dispatching preference given to passenger and freight trains has been raised as a critical issue, but was not explored in previous work. Thus, there is a critical need to update existing relationships to reflect this new operational paradigm.
To quantify the reliability of shared corridors under modern railway operating strategies, the project team will review past relationships and develop appropriate simulation experimental designs incorporating new factor levels to specifically capture important operational changes such as long trains and assigned priority. Using an accepted railway simulation tool, factorial combinations of these factors will be simulated, and results subjected to appropriate statistical analyses to examine changes in Base Train Equivalents over time and between operating conditions. The resulting freight and passenger train performance relationships will be evaluated via case studies to illustrate modern trends in shared corridor reliability.
Analysis of Trends in the Location and Circumstances of Freight Train Handling Derailments
According to the Federal Railroad Administration (FRA) incident-accident database, from 2012 to 2017 there were 101 train handling accidents recorded by the on the mainlines of the four largest Class I railroads in the United States; from 2018 to 2023 there were 177 such train handling accidents, an increase of 75%. These two periods have also marked an increase in the length of freight trains operated in North America, and a decrease in train crew seniority and operating experience in the years following the COVID-19 pandemic, leading both trends to be cited as potential causing factors leading to the observed increase in train handling derailments
over time. However, since correlation does not imply causation, there is a need to further investigate these train handling derailments for possible shifts in their circumstances and locations over time. The nature of these geospatial and circumstantial changes may provide insight into the reasons for the increase in the frequency of train handling derailments, and help inform and optimize industry efforts to mitigate these incidents.
To address this research need, this project will use a combined statistical and geospatial analysis to determine if the location and sizes of trains involved in train handling accidents has changed between the two study periods of 2012-2017 and 2018-2023. Analysis of accident narratives and Geographic Information Systems (GIS) tools will be used to identify how train handling derailment hot spots have shifted between the two periods. Further clues regarding the changing frequency of train handling accidents could be the railroad track topography in the vicinity of the accident locations. Track features such as steep grades, undulating territory, and reverse curves could increase the risk of a train handling accident for longer trains or inexperienced crews. The proposed research will GIS use to overlay the North American Class I freight rail network onto elevation source data to estimate the grade and curvature of track segments in the vicinity of train handling derailments using sinuosity. The distribution of derailments by track topography will be compared to the overall composition of the network to identify if critical conditions are observed to change between the two study periods. Finally, the Advanced Locomotive Technology and Rail Infrastructure Optimization System (ALTRIOS) will be used to conduct train performance simulations over the track topography at each incident location to determine the throttle and/or brake settings leading up the incidents. The magnitude and number of changes in
those settings leading up to the accident location could infer the train handling demands placed on the train crew, and how those demands have changed between the two study periods.
The results from this research can inform freight railroads as to how the locations and circumstances of train handling accidents have changed as trains have become longer, enabling them to more effectively and efficiently deploy mitigation strategies. The proposed spatial analysis methodology could also be adapted and applied to investigate other railroad incident causes to improve overall rail network safety.
Comparative Evaluation of Capital and Spot Tamping Effectiveness through Track Geometry Deterioration Rates
Railroad track deterioration is governed by the interaction between demand factors, such as accumulated traffic and axle loads, and capacity factors including track structure and maintenance practices. Tamping is one of the most widely used maintenance interventions for restoring track geometry, yet it is implemented using different strategies. Capital tamping—also referred to as production or line tamping—treats extended sections of track, while spot tamping targets localized defects in response to emerging geometry issues. These strategies differ in cost, operational impact, and potential influence on long-term track performance. However, the relative effectiveness of these approaches in altering subsequent track geometry deterioration remains insufficiently quantified.
This project proposes the development of a comparative analytical framework to evaluate how different tamping strategies influence track geometry deterioration rates. Using multi-year geometry inspection data collected by Autonomous Track Geometry Measurement Systems (ATGMS) alongside maintenance records from a Class I railroad corridor, the study will identify deterioration cycles and associated maintenance events. Geometry measurements will be aggregated into fixed track segments and analyzed using statistical and signal-processing methods to identify degradation cycles and quantify changes in track condition before and after tamping activities.
The primary focus of the work is to establish a repeatable methodology for quantifying deterioration rates surrounding tamping interventions and comparing the resulting performance trends across different maintenance strategies. The outcome will be a data-driven framework for evaluating tamping effectiveness using longitudinal geometry records.
Such a framework could subsequently be used to investigate how factors such as traffic loading, track structure, maintenance frequency, and intervention type influence deterioration behavior.
Ultimately, this capability supports improved maintenance planning, better alignment of tamping practices with asset management strategies, and more efficient allocation of track maintenance resources.
Advanced Simulation Modeling of Tank Car Safety Performance in Derailments
Train derailments resulting in hazardous materials releases from tank cars can have serious health, environmental, and economic consequences. Given the chaotic and complex nature of train derailments, quantifying the release probability of derailed tank cars is challenging. Statistical quantification of the safety performance of different tank car designs and tank car components in derailments has been conducted using an extensive historical database. However, these statistical probabilities are not a robust metric for new or novel tank car designs or materials because of the lack of data. Instead, analytical or simulation-based approaches are an appropriate alternative for these cases. These require an understanding of the underlying physics of derailment impact events and the dynamic behavior of derailed cars. Information on the
probabilistic impact load environment can be obtained by simulating a series of derailment scenarios that replicate derailment conditions.
An advanced computer model has been developed and partially validated that simulates derailment kinematics enabling estimates of the release probability of tank car components exposed to derailment forces to be estimated. In particular, the model develops probabilistic estimates of the resistance and demand for tank cars and their key components (i.e., tank heads, tank shells, top fittings, and bottom fittings). These models characterize the structural behavior of each tank car component in an impact scenario. They incorporate key elements, such as the derailment-caused impact forces, impact types, impact event characteristics, and tank car properties, to determine the quantities of interest at the level of an impact type affecting a component of a tank car. This model will be further validated and tank car release probabilities estimated using a design of experiments approach for a representative series of derailment scenarios. The validation procedure involves a series of primary and derived metrics to quantify characteristic dynamic behaviors of the train and derailed railcars, such as the extent of the derailment, the spatial dispositions of derailed railcars, their longitudinal and lateral spread, and their misalignment with the track.
Efficacy of 4-Quad Grade Crossing Gates Compared to Other Warning Systems
The goal of this research is to understand and quantify the effect of four-quadrant gates on the safety of highway-rail grade crossings. There are several questions including whether four-quad gates reduce the rate of collisions between highway vehicles and trains compared to other grade crossing warning systems and if so by how much. Specifically, the project will investigate and attempt to quantify the marginal benefit of four-quad versus two-quad gate crossings. Related questions include how do crossing characteristics such as Annual Average Daily Traffic (AADT) or daily train traffic volume interact to affect the efficacy of four-quad versus two-quad gates.
This research will use the FRA’s Highway-Rail Grade Crossing Incident Database (Form 57) and the FRA’s Annual Snapshots taken from the Crossing Inventory Database (Form 71). Use of these databases will provide extensive details about the aforementioned characteristics of grade crossings. Using the data from these sources, both serial and parallel comparative methods will be employed. The serial analysis uses negative binomial regression to compare incidents at a population of crossings before and after they were converted from two-quad to four-quad gates. The parallel analysis uses categorical data analysis to compare the incident rate in two populations of crossings in which we control for Annual Average Daily Traffic (AADT) and train frequency but that differ in their use of two-quad versus four-quad gates. Preliminary results
suggest that there is little difference in the incident rate of crossings equipped with four-quad gates compared to those with two-quad gates calling for more in-depth statistical analysis, and an investigation comparing motor vehicle operator behavior at crossings with these two types of warning systems.