CHAPTER 1
Background and Literature Review

This chapter examines the state of practice regarding derailment loads on transit structures to determine existing gaps in research, which directed the research methodology for this project. The research began with an extensive literature review, which included observing applicable codes and standards in the industry, current technical literature on derailment simulations, the history of train derailments in the United States, and current design practices in light rail transit (LRT) bridge construction.

1.1 Applicable Codes and Standards

Section 3.2.4 (Derailment Load: DE) of the AASHTO LRFD Guide provides general guidelines on derailment loads for the design of aerial structures carrying transit loads, as shown in Figure 4. The guidelines state that the superstructure and substructure members must be designed for simultaneous vertical and horizontal loads caused by a derailment.

The magnitude of the vertical derailment load is the design LRT load, which comprises the weight of the train with all components and the maximum load of seated passengers, operators, and standing passengers (i.e., AW3 condition in rail vehicle terminology), plus a derailment-induced dynamic load allowance, which is specified as 100%. The location of the vertical load is provided by the lateral train excursion distance, which is taken as a minimum of 4 in. to a maximum of 36 in. if guardrails are not present and if the track radius is greater than or equal to 5,000 ft. When the track radius is less than 5,000 ft and guardrails are not present, 8 in. from the barrier wall face or edge of the deck are used. When guardrails are present, the maximum excursion limited by the guardrails is used. For the horizontal load, a prescribed 8 kip/ft force shall be applied at 2 ft above the top of the rail or at the top of the barrier wall, whichever produces the larger force effects. The horizontal load is to be applied normally to the barrier over a 10 ft length. The guidelines do not specify a distinction between various deck construction types and thus are applicable for both ballasted and direct fixation tracks.

The guideline on derailment loads directly follows the subsection outlining dynamic load allowances for live loads, denoted as IM, and the derailment load, denoted as DE, is applied by setting the IM equal to 100%. The dynamic load allowance is defined in the AASHTO LRFD Guide as an increment to be applied to the static wheel load to account for wheel load impact from moving trains. Other cases of IM are 30% for the strength and fatigue limit states and 75% for discontinuities in the superstructure, such as deck joints. The static wheel load is outlined in Section 3.2.1.2 (Design Light Rail Transit Load) of the AASHTO LRFD Guide, obtained from either the LRT-16 load model or the agency-specific light rail live load, whichever generates the larger force effects. LRT-16, which is the notional live load model designated by the AASHTO LRFD Guide, consists of three concentrated loads (i.e., three axles of 34 kips at a constant spacing of 14 ft) and a uniformly distributed load of 0.96 kip/ft. Using a train-related database from

Long Description.

The illustration depicts current guidelines for derailment impact on bridges in the AASHTO LRFD Guide. The left side, labelled a, shows two parallel tracks with guardrails and vertical arrows marked DE acting downward at the rail heads. The curb at the deck edge rises 8 inches when the curve radius is less than 5,000 feet, while for a radius greater than or equal to 5,000 feet, the lateral distance from the rail centerline to the curb face ranges from 4 inches to 3 feet. The right side, labelled b, shows a rail vehicle outline on the track with a horizontal arrow marked D E directed toward the curb at a height of 2 feet above the rail top or at the top of the curb, indicating the point of horizontal load application during a derailment. Both views rest on two piers and depict rail fasteners along the deck.

APTA containing various train weights and axle configurations, the notional live load model was established by 48,256 deterministic light rail loading cases, along with four probability-based examination categories plus a load-enveloping assessment using 660 load cases from 33 trains. The procedure is similar to the approach used to develop the HL-93 truck live load model used to design highway bridges. The AASHTO LRFD Guide commentary compares LRT-16 loading with the AW4 load level (which consists of the weight of the empty train, fully seated passengers, and standees) for the 33 trains, as shown in Figures 5 and 6. The studies indicate that the LRT-16

Figure C3.2.1.2-1—Load enveloping of LRT-16 against 33 Light Rail Trains in AW4: (a) Moment Comparison (205 standees), (b) Shear Comparison (205 standees), (c) Moment Comparison (240 standees), and (d) Shear Comparison (240 standees)

Long Description.

Four graphs from the AASHTO LRFD Guide Specifications for Bridges Carrying Transit Loads display the relationship between span length in feet and structural responses, moment and shear, under 330 load cases and the LRT 16 load model. Graph a plots moment in kip-feet against span length, showing increasing values with span, with LRT 16 values below the envelope formed by the 330 cases. Graph b plots shear in kips with similar trends, where the LRT 16 again falls below the load case range. Graph c replicates moment versus span length for a different set of 330 load cases, maintaining the same LRT 16 position. Graph d replicates shear versus span length behavior. All graphs indicate that LRT 16 is conservative relative to the 330 evaluated load scenarios. Dashed lines represent trend curves, and the open circles denote individual load case results.

Figure C3.2.1.2-2—Response Ratios of LRT-16 against 33 Light Rail Trains in AW4: (a) Moment Ratio (205 standees), (b) Shear Ratio (205 standees), (c) Moment Ratio (240 standees), and (d) Shear Ratio (240 standees)

Long Description.

Four graphs of the AASHTO LRFD Guide Specifications for Bridges Carrying Transit Loads show moment and shear ratios plotted against span lengths up to 300 feet. In graph a, the moment ratio remains below 1.0 across all span lengths, indicating the LRT 16 load model overestimates moments compared to the 330 load cases. Graphs b, c, and d show shear ratios also remain below or near 1.0, with slight increases at longer spans, confirming that the LRT 16 is generally conservative. The dashed horizontal line at 1.0 marks parity, and open circles represent individual load cases across varying span lengths. Overall, the data support using LRT 16 as a safe load model for design.

represents the live load of operational light rail trains in the United States and is therefore appropriate for the design of LRT-carrying bridges.

The 100% derailment vertical impact load from the AASHTO LRFD Guide has been widely adopted by North American transit agencies in their respective design criteria over the last 15 years. In 2002, Dallas Area Rapid Transit (DART) introduced the criteria for a 100% impact factor applied normally to all bridge elements for all axles in a derailed truck [2]. Since then, the DART criteria have been applied to most transit systems in the United States, such as the 2016 Sound Transit Design Criteria Manual [3], the 2016 Valley Metro LRT Design Criteria Manual [4], and the 2009 Regional Transportation District Commuter Rail Design Criteria [5], and Canada, such as the 2005 City of Edmonton LRT Design Guidelines [6] and the 2009 Toronto Transit Commission Design Manual [21]. Older design criteria preceding the 2002 DART criteria that also considered derailment include the 1991 Bay Area Rapid Transit (BART) Structural Design Criteria [7] and the 1998 Pittsburgh Light Rail Design Criteria [8]. The 1991 BART Structural Design Criteria required only the deck slab to be designed for a vertical load equivalent to the static load of a single wheel, with no checks for additional bridge elements, while the 1998 Pittsburgh Light Rail Design Criteria required only a 30% factor be applied to all axles in the train. The evolution of the changes in the definition of derailment load has not been well documented with respect to its technical basis.

“Guidelines for New Development in Proximity to Railway Operations,” published by the Federation of Canadian Municipalities and the Railway Association of Canada, contains

guidelines for designing structures, such as crash walls, for the horizontal impact of derailed passenger train vehicles [22]. The guidelines require consideration of two basic impact cases:

1. Glancing Blow: Two locomotives weighing 148 tonnes each, plus six cars weighing 74 tonnes each, impacting the wall at 10° to the wall.
2. Direct Impact: Single car weighing 74 tonnes impacting the wall at 90° to the wall.

The guidelines require the designer to consider the speed of the derailed train or car impacting the structure to be equal to the specified track speed. The guidelines also allow the designer to account for the plastic energy dissipation of the cars impacting the structure—the maximum deformation may be up to 1 ft per car. The designer may then calculate the design force for the structure as the kinetic energy of the train or car divided by the total plastic deformation of the train or car. For the glancing blow case, the maximum plastic deformation is 8 ft because a total of eight cars are in the train. For the direct impact case, the maximum plastic deformation is 1 ft. The design force is sensitive to the maximum plastic deformation. However, the source of the maximum plastic deformation is unclear. Section 1.2 (Sources) of the guidelines states that information used in the report is derived from “a thorough review of academic literature” and “extensive stakeholder interviews.” Crash energy management (CEM) systems of passenger cars may be designed to have a plastic deformation of up to 1 ft. However, CEM systems are not typically designed for glancing blows. Furthermore, plastic deformation tends to be concentrated at the impact location rather than distributed throughout all cars in the train. Therefore, the assumption of plastic deformation of 8 ft for a glancing blow is likely unrealistic and nonconservative.

1.2 Technical Literature on Derailment Simulations on Bridges

A study of a simplified derailment scenario conducted in 2022 by Lobo and MacNeill indicated that the peak amplification of the live load varied from 300% to 625% depending on the considered vertical train drop heights and bridge stiffnesses [10]. A typical in-service commuter rail transit bridge was selected and modeled as a single-degree-of-freedom (SDOF) system consisting of a rigid deck with a representative mass supported by a spring with a representative bridge stiffness. Three bridge stiffnesses were considered, which corresponded to natural frequencies of 2.0 Hz, 2.5 Hz, and 3.0 Hz, which encompass the range of frequencies seen in U.S. transit bridges. Two drop heights were considered in the analysis, which corresponded to the free fall of a train car from rail to plinth and rail to deck. Both single- and bi-level train cars representative of typical commuter cars in empty, ready-to-run (i.e., AW0) conditions were considered. The authors concluded that because of the high instantaneous impacts, the current code-prescribed derailment impact factor of 2.0 (100%) may not necessarily be conservative and that additional studies are warranted. Further research conducted in 2023 by Catella et al. builds upon the 2022 study conducted by Lobo and MacNeill by examining a range of high-fidelity car structure models of heavy and light rail train cars [11]. For LRT vehicle models, the effects of minimum and maximum operational load conditions (i.e., AW0 and AW3) on derailment impact factors were compared. Dynamic impact factors between 320% and 617% were calculated for a range of LRT configurations, which is consistent with the findings in the 2022 study on single-level and bi-level commuter cars. A simplified approach was also proposed in employing the rigid body kinematics of two colliding bodies to calculate a coefficient of restitution (COR), which showed good agreement with the finite element analysis (FEA). The 2023 paper by Catella et al. further affirms the need for a detailed examination and approach of the derailment forces currently used in North American codes.

Research conducted in 2023 by Lim and Kong [23] focused on wheel–ballast interaction in post-derailment events. An explicit dynamic FEA of derailed trains was performed with a range

of initial velocities, wheel rotational speeds, and wheel–ballast friction coefficients. The characteristics of the track and ballast were applied as slip rate–friction functions. The simulations used the Stribeck curve to define the friction relationship rather than the Coulomb friction model. The simulation results were compared to real-world accidents. The simulations showed a linear reduction in velocity from the initial post-derailment velocity to rest. The results were consistent with an equivalent Coulomb coefficient of friction of approximately 0.2. The simulations were generally consistent with real-world accidents, although the simulations overestimated the longitudinal translation distance—likely because additional obstructions, such as ties, were not included in the simulation. Prior studies conducted in 2021 by Kirkpatrick [26] and in 2006 by Paetsch et al. [27] performed derailment simulations of freight trains like those performed by Lim and Kong in 2023. These earlier simulations used a Coulomb friction model with a median off-track coefficient of friction of 0.6.

1.3 Train Derailments in the United States

Although research on derailment events, specifically for passenger train vehicles, is scarce, Liu et al. conducted a statistical analysis of the causes of major train derailments by observing data from the FRA between 2001 and 2020 [12]. The analysis indicated that broken rails or welds were the leading cause of derailment on main tracks (i.e., excluding yard or siding tracks), accounting for 15.3% of derailments, followed by track geometry (7.3%), car-bearing failure (5.9%), broken car wheels (5.2%), and train handling (4.6%).

The study also compared the accident severity attributed to each derailment cause, as measured by the number of cars derailed per accident, and found that broken rails or welds are not only the most frequent cause of derailment, but they also lead to more severe accidents than average. The highest average severity of the resultant derailments came from cause groups associated with rail defects at bolted joints, other rail and joint defects, and joint bar defects, although the frequency of such occurrences is far lower than the derailment from broken rails or welds. Furthermore, the study separated derailment events by speeds and found that as speeds increased, human factor causes became almost completely absent and instead were replaced with equipment causes. However, broken rails or welds were the leading cause of derailments for all speed ranges.

Cases of derailment on bridges have occurred in the United States, with the most extreme example being the derailment of a freight train on a Union Pacific bridge in Tempe, Arizona (AZ), in June 2020, which resulted in the collapse of several approach spans of the bridge, as detailed in the NTSB report titled “Union Pacific Railroad Derailment with Bridge Strike and Fire” [13]. The structure collapsed within approximately 4 seconds of the derailment, suggesting that the structural failure of the bridge was unlikely to be caused by fire, given the short period between derailment and collapse. Although the 2022 NTSB report did not examine the bridge failure in detail and the cause of the collapse was not established, the vertical impact load of the derailed car is a possible cause of the collapse. More recently, a derailment occurred near a railroad bridge crossing over Interstate 25 (I-25) north of Pueblo, Colorado (CO). While the actual derailment was initiated near a track switch east of the bridge, the derailed cars struck the bridge, causing a partial collapse of the eastern span, and dropped to the interstate below. According to the NTSBʼs preliminary assessment, a broken rail caused the derailment and subsequent bridge collapse; a detailed investigation is ongoing [14].

These two incidents highlight the most severe outcomes of train derailments on bridges. However, several high-profile cases of train derailments in major metro transit lines have occurred in recent years. In January 2024, two New York metro subway trains in Brooklyn derailed in

the span of a week [15], with the second derailment occurring on an elevated trackway. Also in January 2024, an eight-car BART train derailed because of a miscommunication when a manual alignment was performed at the interlocking, which was precipitated by a loss in field communication from the computer system that monitors and manages the track [16][17]. DART also experienced a derailment on its Red Line due to a mechanical failure in October 2023 [18].

1.4 Current Design Practices in LRT Bridge Construction

TCRP Report 155: Track Design Handbook for Light Rail Transit, Second Edition lists the four types of bridge deck construction: open deck construction, embedded deck construction, ballasted deck construction, and direct fixation deck construction [19]. Open deck construction and embedded deck construction are less common and are typically seen in older structures. In open deck construction, timber ties are directly attached to the steel superstructure. The design guidelines from TCRP Report 155 state that this type of construction is not preferred for new or reconstructed bridges and should only be employed when necessary. In embedded deck construction, the rail is embedded within concrete pavement to allow the roadway to be shared between light rail vehicles (LRVs) and other non-rail traffic.

Direct fixation deck construction was developed in the 1960s and is now the standard practice for bridges spanning more than 300 ft and the most common trackform for LRT on aerial structures [19]. Most direct fixation fasteners are supported and anchored to the superstructure on plinths or concrete pads; other methods include cementitious grout pads, booted tie installation, and plinthless direct fixation track installation. Although direct fixation has become commonplace in newer bridge decks, ballasted deck construction is still being used on short-to-moderate-length bridges, generally 300 ft or less.

Section 7.2 of TCRP Report 155 states that “there is no nationally accepted design code that has been developed specifically for light rail transit aerial structures” [19]. In addition, “the older ‘legacy’ rail transit systems (e.g., in Chicago, Philadelphia, and New York) often refer to the AREMA Manual for Railway Engineering for the design of bridges, but the newer systems (e.g., in Atlanta, Washington, D.C., and Baltimore) base their designs on the AASHTO specifications” [19]. In October 2007, however, “the Federal Highway Administration (FHWA) mandated that new and replacement highway bridges that are part of federal-aid funded projects are to be designed according to AASHTO Load and Resistance Factor Design (LRFD) Bridge Design Specifications” [19]. The report also notes that “most light rail transit loads are greater than the HS20 truck load used by the AASHTO specifications, but . . . much less than the Cooper E80 railroad loading cited in the AREMA manual.”

NCHRP Research Report 851: Proposed AASHTO LRFD Bridge Design Specifications for Light Rail Transit Loads discusses current limitations in LRT bridge design and presents a state-of-the-art review and technical investigations into the response of bridges subject to LRT loads [20]. Section 1.1 of NCHRP Research Report 851 states that “due to the absence of standard specifications for light rail transit loading, design of bridge structures is site-specific and largely dependent upon the criteria of individual transit agencies.” It is also noted that some state departments of transportation have material requirements for light rail construction. Because of LRTʼs individualized nature and its dependency on geometric and structural requirements, various bridge constructions can be selected for it. NCHRP Research Report 851 lists typical superstructures for LRT as “prestressed concrete girders, steel plate or box girders, and cast-in-place concrete multi-cell boxes.”

1.5 Existing Gaps in Research

The effects of train derailment on bridges have not been studied extensively, and this fact is reflected in the lack of a clear basis for the 100% derailment impact factor adopted in the AASHTO LRFD Guide. Furthermore, the application of the derailment load is established by setting the IM to 100%, where IM is, in essence, a direct multiplier on the static wheel load. The static wheel load comprises three equal concentrated loads corresponding to the axles according to a notional live load designated as LRT-16 or an agency-specific light rail live load. In a real-life derailment scenario, the wheel load impacting the bridge deck may not be accurately represented by the static wheel load referenced in the AASHTO LRFD Guide since it is evenly distributed across axles.

Catella et al. recently conducted analytical studies of simplified derailment scenarios of LRT vehicles on bridges, studying the validity of the 100% impact factor. However, the existing state of research on derailment impact loads is limited to simple FEA consisting of the bridge structure modeled as a rigid deck supported on an SDOF spring with a representative bridge stiffness. Little to no existing research involves derailment simulations of passenger trains on a more detailed bridge model, which can also allow for variable impact locations, such as near stiffer support points (i.e., piers and abutments).

Furthermore, in previous studies, instantaneous impact force was modeled by simply applying a vertical velocity to the car equivalent to the free fall drop height. Considerations such as capturing the effects of only one truck derailing at a time or prescribing motion to the rail car prior to derailment would more accurately simulate a real-life derailment event. These considerations would also result in horizontal forces being applied to the bridge, which has not been studied in detail and is not explicitly addressed in the current AASHTO LRFD Guide, because the only current requirement is a constant load applied to the barrier, independent of the actual train.