Major Factors Affecting Passenger Train Accident Occurrence and Severity in the United States: Paper Number:

Transcription

1 Lin, Saat and Barkan TRB -0 Major Factors Affecting Passenger Train Accident Occurrence and Severity in the United States: 0 Paper Number: -0 Submitted for Presentation at the th Annual Meeting of the Transportation Research Board and Publication in Transportation Research Record Submission Date: August 0 0 Chen-Yu Lin, Mohd Rapik Saat, and Christopher P. L. Barkan Rail Transportation and Engineering Center Department of Civil and Environmental Engineering University of Illinois at Urbana-Champaign 0 N. Mathews Ave., Urbana, IL, 0 Fax: () - Chen-Yu Lin () - clin@illinois.edu Mohd Rapik Saat (0) - rapik@aar.org Christopher P. L. Barkan () - cbarkan@illinois.edu 0, words + Figures + Tables =, Total words Corresponding Author Current Affiliation: Association of American Railroads. Third St., SW, Washington, DC 00 TRB 0 Annual Meeting

2 Lin, Saat and Barkan TRB -0 0 Abstract Demand for regional and intercity passenger transport in the United States is increasing, resulting in the need to expand rail services and capacity. Railroads are viewed as a promising alternative to highway and air transport because of their ability to provide safe, economical, comfortable, and reliable transport. Passenger train safety has improved over the past two decades; however, faster and more frequent service brings with it greater exposure to potential accidents. Understanding risk in passenger train operations is essential to most efficient resource allocation to further improve safety and reduce the risk of accidents and casualties. This paper presents an analysis of passenger train accidents in the United States from to 0 in order to understand the general trend of passenger train accident rates, quantify the frequency and severity of different accident types, and identify the major factors that cause them. This analysis of train accident provides a foundation for further improvement in passenger train safety and suggests opportunities for future research including shared-use rail corridor risk assessment, train accident precursors, human factor analyses and data mining applications in railroad safety. TRB 0 Annual Meeting

3 Lin, Saat and Barkan TRB -0 Introduction Demand for regional and intercity passenger transport in the United States is increasing, resulting in the need to expand transportation network capacity. Railroads are viewed as a promising alternative to highway and air transport because of their ability to provide safe, economical, comfortable, and reliable transport. All types of passenger rail ridership have been growing in the United States especially in recent years (Figure ). Development of high-speed rail and higherspeed rail further highlight the increase in demand for faster and more frequent passenger rail transportation (, ) TRB 0 Annual Meeting

4 Ridership (Million Unlinked Trips) Ridership (Million Unlinked Trips) Lin, Saat and Barkan TRB Ridership (Million Unlinked Trips) (a) Intercity (Amtrak) (b) Commuter Rail (c) Transit (d) Light Rail Transit 0,00,000,00,000,00,000,00, FIGURE Passenger railroad ridership: (a) intercity (Amtrak), (b) commuter rail, (c) transit and (d) light rail transit from 0 by million unlinked trips (, ) TRB 0 Annual Meeting

5 Lin, Saat and Barkan TRB As service and ridership grow, so do implications for safety, especially with higher speed operation. A number of new or expanded services involve the use of shared trackage, rights-ofway (ROW) and corridors by passenger trains with freight or rail transit systems. The majority of commuter and intercity passenger trains operate on or next to freight railroad corridors (). While these shared-use systems offer passenger rail operations benefits such as lower capital costs, less environmental impact, and easier access to urban cores compared to building a new dedicated line, they also incur some risk such as potential collisions between passenger trains and freight trains () and other risk associated with greater exposure. Understanding these potential aspects in passenger train operations is essential for better resource allocation to improve passenger train safety and reduce the risk of accidents and casualties. Literature Review Research on train accident analyses in the United States has mostly focused on freight train derailments (-), hazardous material releases (-) and grade crossing incidents (-). Few studies have focused on quantitative analysis of U.S. passenger train safety. Much of the research that has been done investigated passenger rail equipment crash energy management systems intended to reduce casualties in a train collision or derailment (-). Lin et al. () conducted a fault tree analysis to identify major factors that could lead to an adjacent track accident on shared passenger and freight rail corridors and developed a semi-quantitative risk assessment model to evaluate the risk (). Internationally, there have been more studies on passenger train accidents. Chen et al. used Associated Rule and other data mining techniques to analyze Chinese passenger train accidents (). Ouyang et al. used System Theoretic Accident Models and Process (STAMP) to analyze a severe railway accident in Jiaoji Railway in China (). Niwa analyzed significant Japanese railway accidents by five major aspects and conducted case studies of several severe accidents (0). Evans used statistical methods to analyze fatal train accidents on Britain s mainline railways (). The author proposed an exponential function to predict the declining trend of train accident rates and applied to other mainline railway systems in Japan, Britain and Europe (- ). Silla and Kallberg studied the development of the railway safety in Finland (). Santos- Reyes and Beard used the Systemic Safety Management System (SSMS) model to analyze two major passenger train accidents in the United Kingdom (-). Britton et al. conducted causal analysis of train derailments in Australia (). These studies are important in providing insights for accident analysis methodologies and results for reference and comparison; however, there are a number of differences in operating practices, rolling stock and organizational structure that affect passenger train safety in the U.S. environment. Further study of passenger train accidents is necessary to understand how to most effectively manage and reduce the risk of U.S. passenger trains. Research Objective This paper presents an analysis of passenger train accidents in the United States from to 0. The objective is to understand the general trend of passenger train accident rates, quantify the frequency and severity of different accident types, and identify the major factors that cause them. TRB 0 Annual Meeting

6 Lin, Saat and Barkan TRB -0 0 Passenger Train Accident Analysis 0 Train accident data from the United States Department of Transportation (U.S. DOT) Federal Railroad Administration (FRA) was used for the analysis (). Railroad accidents/incidents that resulted in monetary loss exceeding a specific threshold must be reported to the Rail Equipment Accident (REA) database maintained by FRA (0). This threshold is periodically adjusted for inflation and is low enough so that only relatively minor incidents are not included. The FRA categorizes train accidents into types (Table ). For the purpose of the analysis described in this paper, these types were consolidated into groups. Incidents of defective pantograph or overhead catenary system (OCS) do not represent the type of safety hazard of principal interest in this research and were excluded. 0 TABLE FRA Train Accident Type Accident/Incident Type Type Code Category in This Paper Derailment Derailment Head-on collision Collision Rear collision Collision Side collision Collision Raking collision Collision Broken-train collision Collision Grade crossing incident Grade Crossing Railroad crossing collision Collision Obstruction Obstruction Explosive-detonation 0 Miscellaneous Fire/violent rupture Miscellaneous Other impact Miscellaneous Others (with description) Miscellaneous Pantograph/OCS N/A Excluded Passenger train accident rate is calculated as number of accidents per million passenger train miles. Over the past 0 years, the overall rate decreased from 0. accidents per million passenger train miles in, to 0. accidents per million passenger train miles in 0 (Table ). However, it is evident from Figure a that the decline fluctuated widely from year to year especially during the first decade of the study period. The overall rate was broken down into the five accident type groups defined above: derailment, collision, grade crossing, obstruction and miscellaneous (Figure b). There was a fairly steady decline in derailments beginning in the mid- 000s. This is consistent with the decreasing trend in freight train derailments that occurred at the same time (). This reduction may be due, in part, to similar factors since most passenger trains in the United States operate on freight railroad tracks. For other accident type groups, no obvious trend was observed. Grade crossing accidents in general have the highest rate among all train accident type groups and demonstrate no obvious downward trend, despite the substantial general decline in grade crossing accidents (, ). TRB 0 Annual Meeting

7 Lin, Saat and Barkan TRB Derailment Collision Grade Crossing Obstruction Miscellaneous Total Year Accident Rate Accident Rate Accident Rate Accident Rate Accident Rate Accident Rate Total ,0 0. TABLE Annual Number of FRA-reportable Mainline Passenger Train Accident by Accident Type, 0 TRB 0 Annual Meeting

8 Accident Rate (Accidents Per Million Passenger Train Miles) Lin, Saat and Barkan TRB -0. Accident Rate (Accidents Per Million Passenger Train Miles) Year (a) Derailment Collision Grade Crossing Obstruction Miscellaneous Year (b) FIGURE FRA-reportable mainline passenger train accident rates, 0 (a) overall rates (b) rates by accident type TRB 0 Annual Meeting

9 Average Casualties Per Accident Lin, Saat and Barkan TRB -0 0 Risk is typically defined as the probability of a particular type of event multiplied by its severity (). In order to identify the types of accident that pose greater threat (i.e. higher probability, severity, or both), accident rate and severity for each type of mainline passenger train accident were plotted in a frequency-severity graph (Figure ). Frequency-severity graphs are a helpful risk visualization tool for train accidents because they enable comparison of the relative frequency and severity of different accident types. They have been used in a number of other railroad accident analyses (0,, ). The graph is divided into four quadrants on the basis of average frequency and severity along each axis. Frequency in this graph is defined as train accident rate. Several different variables were considered to measure passenger train accident severity. The number of railcars derailed has often been used as a proxy variable to measure freight train accident severity (0,,, -,,, ). For passenger trains, casualties are another pertinent variable to measure accident severity (,, 0-). In this paper, casualties are defined as the total number of onboard passenger injuries and fatalities, and were used as the primary severity indicator. Collision Average Frequency: 0. 0 Derailment Average Casualties:.0 Obstruction Miscellaneous Grade Crossing Accidents Per Million Passenger Train Miles FIGURE Frequency-severity graph for FRA-reportable mainline passenger train accident, 0 0 Accident types in the upper right quadrant of Figure are the most likely to pose the greatest risk because they are both more frequent, and more severe, than average. None of the five accident types fell in this quadrant, but derailments and collisions are most likely to result in high-casualty incidents. Together, they accounted for about % of passenger train accidents, but caused about % of total casualties (Table ). Derailments and collisions also cause more damage to rail TRB 0 Annual Meeting

10 Lin, Saat and Barkan TRB -0 0 equipment and infrastructure and are more likely to result in passenger and crew casualties. Although grade crossing incidents are the most common type of accident, they are among the least severe in their consequences to onboard passengers and crew. Therefore, the rest of this paper examines mainline passenger derailments and collisions in more detail TABLE Summary of Frequency, Accident Rate, Casualties and Average Casualties for Different Types of Passenger Train Accidents, 0 Frequency Percentage Average Accident Rate Total Casualties Percentage Average Casualties Grade Crossing 0.% 0.0,00 0.%. Obstruction.% 0.0.% 0. Derailment.0% 0.,0 0.%. Collision.% 0.0.%. Miscellaneous 0.% % 0. Total,0 00.0% 0., 00.0%. Passenger Train Derailment and Collision Accident Cause Analysis To further understand which factors contributed the most to passenger train derailments and collisions, a more detailed causal analysis of the five accident types was conducted. FRA trainaccident cause codes are hierarchically organized and categorized into major cause groups - track, equipment, human factors, signal, and miscellaneous (0). Each of these major cause groups has subgroups that include individual cause codes of related causes. In this paper, alternative FRA subgroups developed by Arthur D. Little (ADL) and the Association of American Railroads (AAR) are used in which similar cause codes were grouped based on experts opinion (). ADL s groupings enable greater resolution for certain train accident causes. For example, FRA combines broken rails, joint bars and rail anchors in the same subgroup, whereas the ADL grouping distinguishes between broken rail and joint bar defects (). The frequency and severity graph of accidents due to the major accident cause groups was plotted (Figure ). As in Figure, the graph is divided into four quadrants to enable comparison of the relative frequency and severity of the different cause groups. The human factors cause group had above-average frequency and was the most severe in terms of average casualties. It accounted for 0% of the total derailments and collisions but.% of the total casualties (Table ). Track, Roadbed, and Structures accidents were more frequent than human factors, but less severe (.% of the total derailments and collisions and.% of the total casualties). Both human factors and track, roadbed, and structure-related accident causes consistently represented the most frequent and severe accident cause groups, together accounting for a total of.% of derailments and collisions, and.% of casualties; therefore they were analyzed in more detail. TRB 0 Annual Meeting

11 Average Casualties Per Accident Lin, Saat and Barkan TRB Buckled Track Non-Traffic, Weather Causes Average Frequency:. Train Operation Human Factors Track, Roadbed and Structure Miscellaneous Mechanical and Electrical Factors Signal and Communication Joint Bar Defects 0 Misc. Human Factors 0 Train Speed Failure to Obey/Display Signals 0 0 Mainline Rules Broken Rails or Welds Average Casualties:.0 Track Geometry (excl. Wide Gauge) Turnout Defects - Switches Wide Gauge Number of Accidents FIGURE Frequency and severity graph of mainline passenger derailments and collisions, -0, by accident cause category with average casualties 0 TABLE Summary of Frequency, Accident Rate, Casualties and Average Casualties of Mainline Passenger Derailments and Collisions, 0, by Accident Cause Category Frequency Percentage Average Rate Total Casualties Percentage Train Operation Human Factor 0.0% 0.0,.%. Track, Roadbed, and Structure.% 0.0.%. Miscellaneous.% 0.0.%. Mechanical and Electrical Factors.% % 0. Signal and Communication.% % 0.0 Total 00.0% 0., 00.0%. Average Casualties In order to identify trends in specific accident causes, the five-year moving average of combined derailment and collision rate was broken down by accident cause category (Figure ). Infrastructure (Track, Roadbed and Structure) and human factor categories were consistently the most frequent accident cause categories over the 0-year study period, with infrastructure causes being higher for every -year interval except 00. TRB 0 Annual Meeting

12 Accident Rate (Accidents Per Million Passenger Train Miles) Lin, Saat and Barkan TRB Total Infrastructure Human Factor Equipment Signal Miscellaneous Five Year Moving Average Period FIGURE Five-year moving average of combined mainline passenger train derailment and collision rate, 0, by accident cause category The accident cause groups were further analyzed by preparing a frequency and severity graph of the more detailed accident cause subgroups (Figure ). Each data point represents one accident cause subgroup. Data points with the same color and shape indicate that these accident cause subgroups are in the same accident cause category. In terms of average casualties, four accident cause subgroups were in the upper right quadrant, and thus most likely to pose the greatest risk due to their high frequency and severity. All of them are human factor accident causes: Failure to Display/Obey Signals (Human Factors) Train Speed (Human Factors) Miscellaneous Human Factors (Human Factors) Mainline Rules (Human Factors) TRB 0 Annual Meeting

13 Average Casualties Per Accident Lin, Saat and Barkan TRB Buckled Track Non-Traffic, Weather Causes Average Frequency:. Train Operation Human Factors Track, Roadbed and Structure Miscellaneous Mechanical and Electrical Factors Signal and Communication Joint Bar Defects 0 Misc. Human Factors 0 Train Speed Failure to Obey/Display Signals Mainline Rules Broken Rails or Welds Average Casualties:.0 Track Geometry (excl. Wide Gauge) Turnout Defects - Switches Wide Gauge Number of Accidents FIGURE Frequency and severity graph of mainline passenger derailments and collisions, -0, by accident cause subgroups with average casualties These four subgroups accounted for 0.% of the total mainline passenger derailments and collisions but.% of total casualties (Table ). Among all the subgroups identified in the top-right quadrant, Failure to Display/Obey Signals, had the highest average casualties per accident, followed by Mainline Rules, Train Speed and Misc. Human Factors. Irrespective of quadrant, the five most frequent accident-cause subgroups were: Turnout Defect Switches, Failure to Obey/Display Signals, Wide Gauge, Use of Switches and Other Miscellaneous. Combined they accounted for.% of total derailments and collisions and.% of total casualties. Two of the top five most frequent accident cause subgroups were infrastructure related, two of them were human factors and one of them was miscellaneous. TRB 0 Annual Meeting

14 Lin, Saat and Barkan TRB -0 TABLE Summary of Frequency, Accident Rate, Casualties and Average Casualties of Mainline Passenger Derailments and Collisions, 0, by Accident Cause Subgroups Cause Subgroup Code Cause Subgroup Description Accidents Frequency Per Million Casualties Cars Derailed Number Percentage Train Miles Number Percentage Average Number Percentage Average 0T Turnout Defects - Switches.% 0.00.%..%.0 0H Failure to Obey/Display Signals.% %..0%. 0T Wide Gauge.% 0.0.%..%. H Use of Switches 0.% % 0..%. 0M Other Miscellaneous 0.% 0.00.%..%. 0T Track Geometry (excl. Wide Gauge).% 0.00.%..%. E Loco Trucks/Bearings/Wheels.% 0.00.%..%. 0T Broken Rails or Welds.% 0.00.%..%. 0H Mainline Rules.% %..0%. 0H Train Speed.% %..%. 0M Obstructions.% % 0..0%. E Other Wheel Defects (Car) 0.% % 0..%. H Misc. Human Factors 0.% 0.00.%..%. E All Other Car Defects.0% % 0. 0.%. T Misc. Track and Structure Defects.% % 0..%. 0E Centerplate/Carbody Defects (Car).% % 0. 0.% 0. 0S Signal Failures.% % 0..0%. E Other Axle/Journal Defects (Car).% % %. 0H Handbrake Operations.% %..%. E All Other Locomotive Defects.0% % %. 0H Switching Rules.0% % %.0 0M Lading Problems.0% % % 0.0 0T Non-Traffic, Weather Causes.0% 0.00.%..%.0 0T Buckled Track.0% 0.00.%.0.%. 0T Rail Defects at Bolted Joint.0% % %. 0T Joint Bar Defects.0% 0.00.%..%. E TOFC/COFC Defects 0.% % %.0 0M Track-Train Interaction 0.% % 0. 0.%. 0T Roadbed Defects 0.% %..0%.0 0T Other Rail and Joint Defects 0.% %.0 0.%.0 0E Coupler Defects (Car) 0.% % %.0 0E Sidebearing, Suspension Defects (Car) 0.% % %.0 E Stiff Truck (Car) 0.% % %.0 0H Employee Physical Condition 0.% %.0.%.0 T Turnout Defects - Frogs 0.% %.0 0.%.0 0E Air Hose Defect (Car) 0 0.0% % % 0.0 0E Brake Rigging Defect (Car) 0 0.0% % % 0.0 0E Handbrake Defects (Car) 0 0.0% % % 0.0 0E UDE (Car or Loco) 0 0.0% % % 0.0 0E Other Brake Defect (Car) 0 0.0% % % 0.0 0E Truck Structure Defects (Car) 0 0.0% % % 0.0 0E Bearing Failure (Car) 0 0.0% % % 0.0 E Broken Wheels (Car) 0 0.0% % % 0.0 E Loco Electrical and Fires 0 0.0% % % 0.0 0E Track/Train Interaction (Hunting) (Car) 0 0.0% % % 0.0 E Current Collection Equipment (Loco) 0 0.0% % % 0.0 0H Brake Operation (Main Line) 0 0.0% % % 0.0 0H Brake Operations (Other) 0 0.0% % % 0.0 0H Radio Communications Error 0 0.0% % % 0.0 0H Train Handling (excl. Brakes) 0 0.0% % % 0.0 0M Grade Crossing Collisions 0 0.0% % % 0.0 Total 00.0% 0.0, 00.0%. 00.0%. TRB 0 Annual Meeting

15 Lin, Saat and Barkan TRB -0 0 Effect of Speed on Passenger Train Derailment and Collision Cause Previous research has found that the speed of a train at the time of derailment was positively correlated with with derailment severity (0, -, -, 0). Previous research has also found an inverse relationship between FRA track class and freight train derailment rate (,, ). Figure shows the number and percentage of mainline passenger train derailments and collisions by speed and accident cause category. The majority of train accidents about % occurred below 0 mph. This may be related to the high incidence of defective-turnout-caused derailments. Turnouts are found at stations, terminals, and the ends of sidings where trains are likely to slow down due to speed restrictions, scheduled stops or meet/pass activities. Infrastructure related accidents occurred in almost all speed ranges and had the highest percentage except the 0 mph+ category. No specific trends were found for human-factorcaused and equipment-caused accidents; however, the three accidents that occurred with train speed greater than 00 mph were human factor and equipment caused. TRB 0 Annual Meeting

16 Percentage of Accident Within Speed Range Lin, Saat and Barkan TRB -0 Number of Accident Within Speed Range % 0% 0% 0% Accident Speed (mph) (a) Infrastructure Human Factor Equipment Signal Miscellaneous Infrastructure Human Factor Equipment Signal Miscellaneous 0% 0% 0% 0% 0% 0% 0% Accident Speed (mph) (b) FIGURE Number (a) and percentage (b) of mainline passenger train derailments and collisions by speed and accident cause category, 0 TRB 0 Annual Meeting

17 Lin, Saat and Barkan TRB To further understand what caused a derailment or collision at different speeds, Table shows the number of mainline passenger train derailments and collisions by accident cause subgroup in different speed ranges. In the 0-0 mph range, Turnout Defects Switches was the top accident cause subgroup, consistent with our previous inference regarding low speed accident causes. In the -0 mph and -0 mph ranges, Failure to Obey/Display Signals was most frequent. In the -0 and 0-00 ranges, some equipment-related accident cause subgroups, namely All Other Car Defects and Other Wheel Defects (Car) were the top causes. The three accidents in which speed was above 00 mph were in the following three subgroups: All Other Locomotive Defects, Centerplate/Carbody Defect (Car) and Misc. Human Factors. Summaries of these three accidents are as follows (in the order of accident date):. April th, 00. Amtrak train was side-swiped by an improperly secured freight locomotive door from a freight train on the adjacent track (accident type: raking collision; Amtrak train speed: 0 mph, no cars derailed; no casualties; accident cause sub group: All Other Locomotive Defects). January th, 00. Amtrak train was side-swiped by an improperly secured freight car door from a freight train on the adjacent track (accident type: raking collision; Amtrak train speed: 0 mph, no cars derailed, no casualties, accident cause subgroup: Centerplate/Carbody Defect (Car)). May th, 0. Amtrak train derailed at 0 mph, resulting in locomotive and passenger cars derailed; casualties; accident cause subgroup: Misc. Human Factors () TABLE Most Frequent Accident Cause Subgroups of Mainline Passenger Train Derailment and Collision, 0, by Accident Speed Ranges Top Accident Cause Subgroups Ranking by Frequency * Failure to Failure to All Other Car Other Wheel Obey/Display Obey/Display Defects Defects (Car) Signals Signals Turnout Defects - Switches Failure to Obey/Display Signals Loco Trucks/Bearings /Wheels Broken Rails or Welds Use of Switches Wide Gauge Wide Gauge Wide Gauge Other Miscellaneous Turnout Defects - Switches Mainline Rules Other Miscellaneous Obstructions * There were only three accidents where train speed was greater than 00 mph. Accident Speed (mph) Joint Bar Defects Broken Rails or Welds Misc. Human Factors Track Geometry (excl. Wide Gauge) Track Geometry (excl. Wide Gauge) Train Speed Handbrake Operations Non-Traffic, Weather Causes Misc. Human Factors All Other Locomotive Defects Centerplate/Carb ody Defects (Car) A scatterplot of train accident speed versus casualties was prepared (Figure ). A simple linear regression indicates a positive relationship between train accident speed and casualties (P-value TRB 0 Annual Meeting

18 Casualties Lin, Saat and Barkan TRB -0 0 = 0.0; R = 0.); however, the coefficient of determination is low, meaning there are other factors also affecting severity. One example is type of accident. A raking collision between an object from a freight train and the side of a passenger train may not result in large number of casualties, even at high speed ( occurrences; average train speed =. mph; average casualties = 0.). On the other hand, a head-on collision may result in severe casualties at relative lower speed ( occurrences; average train speed =. mph; average casualties =.). Other factors include number of passengers onboard, curvature and grade of the track and location of accident such as on a bridge or in a tunnel. A positive relationship between train accident speed and casualties is reasonable because when an accident occurs at higher speed, more kinetic energy is involved and therefore there is a higher probability of greater consequences Train Accident Speed (mph) FIGURE Scatterplot of casualties and train accident speed DISCUSSION In this paper we analyze U.S. mainline passenger train accidents in the 0-year period from to 0 and identify major accident types and causes. The paper also associated the effect of train speed with accident frequency, severity and accident causes. These findings provide a basis for future improvement in passenger train safety. Based on these results, several directions for future research are discussed. Adjacent Track Accidents on Shared-Use Rail Corridors TRB 0 Annual Meeting

19 Lin, Saat and Barkan TRB With the development of high-speed rail as well as continuing improvement in the conventional passenger rail system in the United States, it is expected that there will be more shared-use rail corridors and more mixed passenger and freight train operations. A consequent safety issue on these shared-use rail corridors is adjacent track accident risk (). Adjacent track accidents are a type of accident in which one train derails and intrudes upon adjacent tracks, and then is struck by trains on those tracks. With more trains operating on a corridor, the probability of train interactions also increases, meaning that if a train derails and intrudes onto an adjacent track, there is a greater chance that another train will be present or approaching on the adjacent track. Furthermore, higher speed of train operations on shared-use rail corridors means the potential consequences of an accident are also greater. Focusing on addressing the risk of adjacent track accidents will help improve our understanding of this risk and lead to more effective risk reduction strategies. Accident Precursors As safety continues to improve in the railroad system, statistical analyses to reliably estimate risk will become more challenging due to the smaller empirical basis for analysis (). To address this, accident precursors must be considered. An Accident Precursor is defined by National Aeronautics and Space Administration (NASA) as: an anomaly that signals the potential for more severe consequences that may occur in the future, due to causes that are discernible from its occurrence today (). An example of an accident precursor in a railroad system is locomotive engineer over-running a stop signal, but without any further consequence such as a collision or derailment. Train accidents are a subset of accident precursors, meaning that under certain conditions, accident precursors will result in train accidents, but most will not. Analyzing accident precursors provides more data and consequently more robust predictive risk analysis. Studying accident precursors also allows researchers to identify preventive measures that mitigate risk at the precursor event level, and potentially further reduce the occurrence of train accidents. For example, if a preventive measure can effectively reduce the probability of an engineer overrunning a stop signal, it can also reduce the probability of a train accident caused by Failure to Obey/Display Signal. Positive Train Control for passenger train operation is an example of a preventive measure that will prevent this type of precursor event, as well as accidents associated with this cause, and certain others as well. Accident precursor analysis has been implemented in the United Kingdom (), and in the United States. The Confidential Close Call Reporting System has been implemented by the FRA to collect close call (or near misses ) data () and these data are a good candidate for train accident precursor analysis. Human Factors Analysis Train Operation Human Factors were identified as the most frequent and severe passenger train accident cause category. Consequently addressing them will be critical to the success of further passenger train risk reduction efforts. Railroad human factor research encompasses a wide spectrum of topics including human fatigue in train operation, ergonomics, and human performance in the train control system. In Europe, Signal Passed at Danger (SPAD), which is similar to the Failure to Obey/Display Signal cause subgroup in the United States, is a critical TRB 0 Annual Meeting

20 Lin, Saat and Barkan TRB human factor topic and has drawn substantial attention as an important source of the risk (0, ). Employee fatigue is another aspect of human factors risk and has contributed to several severe passenger train accidents. Overall, there are opportunities and potential to reduce passenger train derailments and collision by addressing human factors. Data Mining Applications in Railroad Safety Improvement Expanded automated data collection systems, combined with rapid advances in data mining technology, means that new methodologies are available for rail safety analyses. These include associated rules (), STAMP (), and Maximal Information Coefficient (MIC) (). Data mining techniques can be implemented to increase risk model accuracy and to handle complex effects from multiple (and perhaps correlated) influencing factors CONCLUSION This paper presents the results of a study to identify the most important factors contributing to the risk of passenger train accidents. Derailments and collisions were identified as the most potentially significant train accident types, while human factors accidents and track failures were the primary causes of those accidents. Accident causes related to human factors and train operations such as train speed violations and failure to obey signals have high risk. High-risk infrastructure-related factors include track geometry defects and broken rails or welds. Most passenger train derailments and collisions occurred at lower speed. This analysis of train accident causes is important for rational allocation of resources to reduce accident occurrence and consequences and provides a foundation for further improvement in passenger train safety. The paper also suggests opportunities for future research including shared-use rail corridor risk assessment, train accident precursors, human factor analyses and data mining applications in railroad safety. ACKNOWLEDGEMENTS This research was supported by the National University Rail (NURail) Center, a U.S. DOT OST- R University Transportation Center. The views expressed in this paper do not necessarily reflect the views of the Association of American Railroads, the second author s current employer. REFERENCES. Federal Railroad Administration. High Speed Rail Strategic Plan. United States Department of Transportation, Washington D.C., 00.. Federal Railroad Administration. High Speed Intercity Passenger Rail (HSIPR) Program. Accessed June 0.. American Public Transportation Association. Ridership Report Archives. Accessed November 0.. National Railroad Passenger Cooperation. Amtrak National Facts Accessed December 0. TRB 0 Annual Meeting

21 Lin, Saat and Barkan TRB Brod, D., Metcalf, A.E. and Kraft, E. Web-Based Screening Tool for Shared-Use Rail Corridors. Publication NCFRP Report No.. Transportation Research Board of the National Academies, Washington, D.C., 0.. Saat, M.R. and Barkan, C.P.L. Investigating Technical Challenges and Research Needs Related to Shared Corridors for High-Speed Passenger and Railroad Freight Operations. Publication DOT/FRA/ORD--. United States Department of Transportation, Washington, D.C., 0.. Birk, A.M., Anderson, R.J. and Coppens, A.J. A computer simulation of a derailment accident: Parts I model basis. Journal of Hazardous Materials, Vol., 0, pp... Birk, A.M., Anderson, R.J. and Coppens, A.J. A computer simulation of a derailment accident: Parts II sample simulation. Journal of Hazardous Materials, 0, Vol., pp... Dennis, S.M. Changes in railroad track accident rates. Transportation Quarterly, Vol., Issue, 00, pp. 0. Barkan, C.P.L., Dick, C.T. and Anderson, R.T. Railroad derailment factors affecting hazardous materials transportation risk. Transportation Research Record: Journal of Transportation Research Board, Vol., 00, pp... Anderson, R.T. and Barkan, C.P.L. Railroad accident rates for use in transportation risk analysis. Transportation Research Record: Journal of Transportation Research Board, Vol., 00, pp... Liu, X., Barkan, C.P.L. and Saat, M.R. Analysis of derailments by accident cause: evaluating railroad track upgrades to reduce transportation risk. Transportation Research Record: Journal of Transportation Research Board, Vol., 0, pp... Liu, X., Saat, M.R. and Barkan, C.P.L. Analysis of causes of major train derailment and their effect on accident rates. Transportation Research Record: Journal of Transportation Research Board, Vol., 0, pp... Liu, X., Saat, M.R. and Barkan, C.P.L. Analysis of U.S. freight-train derailment severity using zero-truncated negative binomial regression and quantile regression. Accident Analysis & Prevention, Vol., 0, pp... Liu, X. Statistical temporal analysis of freight train derailments rates in the United States: 000 to 0. Transportation Research Record: Journal of Transportation Research Board, Vol., 0, pp... Nayak, P.R., Rosenfield, D.B. and Hagopian, J.H. Event Probabilities and Impact Zones for Hazardous Materials Accidents on Railroads. Publication DOT/FRA/ORD-/0. United States Department of Transportation, Washington, D.C.,.. Saccomanno, F.F. and El-Hage, S.M. Minimizing derailments of railcars carrying dangerous commodities through effective marshaling strategies. Transportation Research Record: Journal of Transportation Research Board, Vol.,, pp.. TRB 0 Annual Meeting

22 Lin, Saat and Barkan TRB Saccomanno, F.F. and El-Hage, S.M. Establishing derailment profile by position for corridor shipment of dangerous goods. Canadian Journal of Civil Engineering, Vol., Issue,, pp... Kawprasert, A. and Barkan, C.P.L. Effect of route rationalization on hazardous materials transportation risk. Transportation Research Record: Journal of Transportation Research Board, Vol. 0, 00, pp.. 0. Bagheri, M., Saccomanno, F.F., Chenouri, S. and Fu, L. Reducing the threat of in-transit derailments involving dangerous goods through effective placement along the train consist. Accident Analysis & Prevention, Vol., Issue, 0, pp. 0.. Liu, X., Saat, M.R. and Barkan, C.P.L. Integrated risk reduction framework to improve railway hazardous materials transportation safety. Journal of Hazardous Materials, Vol. 0, 0, pp. 0.. Liu, X., Saat, M.R. and Barkan, C.P.L. Probability analysis of multiple-tank-car release incidents in railway hazardous materials transportation. Journal of Hazardous Materials, Vol., 0, pp... Liu, X. and Hong, Y. Analysis of railroad tank car releases using a generalized binomial model. Accident Analysis & Prevention, Vol., 0, pp. 0.. Liu, X. Optimizing rail defect inspection frequency to reduce the risk of hazardous materials transportation by rail. Journal of Loss Prevention in the Process Industries, Vol., 0, pp... Benekohal, R.F. and Elzohairy, Y.M. A new formula for prioritizing railroad crossings for safety improvement. In: Proceedings of the Institution of Transportation Engineers Annual Meeting, Chicago, IL, 00.. Austin, R.D. and Carson, J. L. An alternative accident prediction model for highway-rail interfaces. Accident Analysis & Prevention, Vol., Issue, 00, pp... Saccomanno, F.F., Fu, L. and Miranda-Moreno, L.F. Risk-based model for identifying highway-rail grade crossing blackspots. Transportation Research Record: Journal of Transportation Research Board, Vol., 00, pp... Mok, S.C. and Savage, I. Why has safety improved at rail-highway grade crossings? Risk Analysis, Vol., Issue, 00, pp... Saccomanno, F.F., Park, P.Y.J. and Fu, L. Estimating countermeasure effects for reducing collisions at highway-rail grade crossings. Accident Analysis & Prevention, Vol., Issue, 00, pp Chadwick, S. G., Zhou, N. and Saat, M.R. Highway-rail grade crossing safety challenges for shared operations of high-speed passenger and heavy freight in the U.S. Safety Science, Vol., 0, pp... Williams, T., Abello, J., Betak, J. and Desimore, D. Using data visualization to analyze grade crossing accidents. In Proceedings of the 0 Joint Rail Conference, San Jose, CA, 0. TRB 0 Annual Meeting

23 Lin, Saat and Barkan TRB Williams, T., Nelson, C. and Betak, J. Applying topic modeling to railroad grade crossing accident report text. In Proceedings of the 0 Joint Rail Conference, San Jose, CA, 0.. Tyrell, D.C. US rail equipment crashworthiness standards. Proceedings of the Institution of Mechanical Engineers, Part F: Journal of Rail and Rapid Transit, Volume, Issue, 00, pp. 0.. Tyrell, D.C. Rail passenger equipment accidents and the evaluation of crashworthiness strategies. Proceedings of the Institution of Mechanical Engineers, Part F: Journal of Rail and Rapid Transit, Volume, Issue, 00, pp... Tyrell, D.C. and Perlman, A.B. Evaluation of rail passenger equipment crashworthiness strategies. Transportation Research Record: Journal of Transportation Research Board, Vol., 00, pp... Lin, C.Y., Saat, M.R. and Barkan, C.P.L. Fault tree analysis of adjacent track accidents on shared-use rail corridors. Transportation Research Record: Journal of Transportation Research Board, Vol., 0, pp... Lin, C.Y. and Saat, M.R. Semi-quantitative risk assessment of adjacent track accidents on shared-use rail corridors. In: Proceedings of the 0 Joint Rail Conferences, Colorado Springs, CO, 0.. Chen, D., Xu, C. and Ni, S. Data mining on Chinese train accidents to derive associated rules. Proceedings of the Institution of Mechanical Engineers, Part F: Journal of Rail and Rapid Transit, Volume, Issue, 0, pp... Ouyang, M., Hong, L., Yu, M.H. and Fei, Q. STAMP-based analysis on the railway accident and accident spreading: Taking the China-Jiaoji railway accident for example. Safety Science, Vol., Issue, 00, pp.. 0. Niwa, Y. A proposal for a new accident analysis method and its application to a catastrophic railway accident in Japan. Cognition, Technology & Work, Vol., Issue, 00, pp. 0.. Evans, A.W. Fatal train accidents on Britain s mainline railway. Journal of the Royal Statistical Society. Series A (Statistics in Society), Vol., No, 000, pp... Evans, A.W. Rail safety and rail privatisation in Britain. Accident Analysis & Prevention, Vol., Issue, 00, pp. 0.. Evans, A.W. Rail safety and rail privatisation in Japan. Accident Analysis & Prevention, Vol., Issue, 00, pp. 0.. Evans, A.W. Fatal train accidents on Europe s railways: Accident Analysis & Prevention, Vol., Issue, 0, pp. 0.. Silla, A. and Kallberg, V.P. The development of railway safety in Finland. Accident Analysis & Prevention, Vol., 0, pp... Santos-Reyes, J. and Beard, A.N. A systemic analysis of the Paddington railway accident. Proceedings of the Institution of Mechanical Engineers, Part F: Journal of Rail and Rapid Transit, Volume 0, Issue, 00, pp.. TRB 0 Annual Meeting

24 Lin, Saat and Barkan TRB Santos-Reyes, J. and Beard, A.N. A systemic analysis of the Edge Hill railway accident. Accident Analysis & Prevention, Vol., Issue, 00, pp... Britton, M.A., Asnaashari, S. and Read, G.J.M. Analysis of train derailment cause and outcome in Victoria, Australia, between 00 and 0: Implications for regulation. Journal of Transportation Safety & Security, Vol., Issue, 0, pp... Federal Railroad Administration. Download Data on Demand. U.S. Department of Transportation, Washington, D.C. Accessed July st, Federal Railroad Administration. FRA Guide for Preparing Accident/Incident Reports. U.S. Department of Transportation, Washington, D.C., 0.. Federal Railroad Administration. In-Depth Data Analysis of Grade Crossing Accidents Resulting in Injuries and Fatalities. Publication DOT/FRA/ORD--0. United States Department of Transportation, Washington, D.C., 0.. Elvik, R. and Voll, N.G. Challenges of improving safety in very safe transport systems. Safety Sciences, Vol., 0, pp... Wang, B. Z., Barkan, C.P.L., Liu, X. and Saat, M.R. Trends in U.S. freight train accident causes and rates A quantitative approach (working paper).. Schafer II, D.H. and Barkan, C.P.L. Relationship between train length and accident causes and rates. Transportation Research Record: Journal of Transportation Research Board, Vol. 0, 00, pp... Arthur D. Little, Inc. (ADL). Risk Assessment for the Transportation of Hazardous Materials by Rail, Supplementary Report: Railroad Accident Rate and Risk Reduction Option Effectiveness Analysis and Data (Second Revision). Cambridge, Massachusetts,.. Derailment of Amtrak Passenger Train Philadelphia, Pennsylvania May, 0. National Transportation Safety Board. Accessed June, 0.. NASA Accident Precursor Analysis Handbook. National Aeronautics and Space Administration. Accessed June, 0.. Prospective Study into Harmonised Train Accident Precursors Analysis and Management. The European Railway Agency. Register/Documents/PPR%0Report.pdf Accessed June, 0.. Close Call Reporting. Federal Railroad Administration. Accessed June, Pasquini, A., Rizzo, A. and Save., L. A methodology for the analysis of SPAD. Safety Science, Vol., Issue, 00, pp.. TRB 0 Annual Meeting

25 Lin, Saat and Barkan TRB -0. Wilson, J.R. and Norris, B.J. Rail human factors: Past, present and future. Applied Ergonomics, Vol., Issue., 00, pp. 0.. Shao, F. and Li, K. A graph model for preventing railway accidents based on the maximal information coefficient. International Journal of Modern Physics B, Vol., Issue, 0, 000 ( pages). TRB 0 Annual Meeting