CONDITIONAL PROBABILITY OF RELEASE OF HAZARDOUS MATERIALS FROM RAILROAD TANK CARS USING BAYESIAN NETWORKS. Emmanuel Nii Martey

Transcription

1 CONDITIONAL PROBABILITY OF RELEASE OF HAZARDOUS MATERIALS FROM RAILROAD TANK CARS USING BAYESIAN NETWORKS by Emmanuel Nii Martey A thesis submitted to the Faculty of the University of Delaware in partial fulfillment of the requirements for the degree of Master of Civil Engineering Fall Emmanuel Nii Martey All Rights Reserved

2 ProQuest Number: All rights reserved INFORMATION TO ALL USERS The quality of this reproduction is dependent upon the quality of the copy submitted. In the unlikely event that the author did not send a complete manuscript and there are missing pages, these will be noted. Also, if material had to be removed, a note will indicate the deletion. ProQuest Published by ProQuest LLC (2016). Copyright of the Dissertation is held by the Author. All rights reserved. This work is protected against unauthorized copying under Title 17, United States Code Microform Edition ProQuest LLC. ProQuest LLC. 789 East Eisenhower Parkway P.O. Box 1346 Ann Arbor, MI

3 CONDITIONAL PROBABILITY OF RELEASE OF HAZARDOUS MATERIALS FROM RAILROAD TANK CARS USING BAYESIAN NETWORKS by Emmanuel Nii Martey Approved: Nii Attoh-Okine, Ph.D. Professor in charge of thesis on behalf of the Advisory Committee Approved: Harry W. Shenton III, Ph.D. Chair of the Department of Civil and Environmental Engineering Approved: Babatunde A. Ogunnaike, Ph.D. Dean of the College of Engineering Approved: Ann L. Ardis, Ph.D. Interim Vice Provost for Graduate and Professional Education

4 ACKNOWLEDGMENTS I thank the Almighty God for how far He has brought me. I thank Him for His mercies and guidance throughout the entire period of my work. My sincerest gratitude goes to my graduate advisor, Prof. Attoh-Okine for his patience, support and encouragement throughout my study. I also wish to thank Offei Amanor Adarkwa who has completed his PhD program and Silvia Galvan Nunez, a PhD candidate at the University of Delaware, for their help and contributions to my work. I also wish to thank Chris Reoli for her invaluable assistance. Finally, I wish to thank my family and all my loved ones who have encouraged and supported me throughout my graduate study. iii

5 TABLE OF CONTENTS LIST OF TABLES... vi LIST OF FIGURES... viii ABSTRACT... x Chapter 1 INTRODUCTION General Statement of Problem Objective of Research Research Framework Organization of the Thesis BACKGROUND AND LITERATURE REVIEW History of the Tank Car Tank Car Standards Legacy DOT 111 Tank Cars CPC 1232 Tank Cars DOT 117 Tank Cars Tank Car Design Improvement Train Accidents Involving Tank Car Release Transportation of Crude Oil by Rail Transportation of Ethanol by Rail CPR MODELS General Hazmat Release Risk Logistic Regression CPR Models Multiple-Car Release Probability Methodology iv

6 Total Number of Cars Derailed, P (X K) Tank Car Derailment, P (X D X) Tank Car Release, P (X R X D ) Monte Carlo Probabilistic CPR Models Non-CPR Models Methodology BAYESIAN NETWORKS General Bayes Theorem Joint Probability Distribution Conditional Independence Bayesian Inference Inference Algorithms Exact Inference Algorithms Junction Trees Elimination Trees Shenoy-Shafer Method Approximate Inference Algorithms DATA ANALYSIS AND RESULTS Formulation and Sources of Data Analysis Results CONCLUSION AND RECOMMENDATIONS General Conclusion Recommendations REFERENCES Appendix A PERMISSION LETTERS v

7 LIST OF TABLES Table 1: Table 2: Table 3: Timeline for the Retrofit of Affected Tank Cars for Use in North American HHFTs (USDOT, 2015a) List of recent train derailments in North America occurring over the last quarter of a century (Jeong, 2009; Stancil, 2012; Stephens, 2014; El-Sibaie, 2014; US DOT, 2015b; FRA 2015; TSB, 2015) List of recent train derailments along with key information on each incident (Stephens 2014, El Sibaie 2014, FRA 2015) Table 4: Random Variables and Assumed Distributions (Jeong, 2009) Table 5: Mean and Standard Deviation for Random Variables (Jeong, 2009) Table 6: Table 7: Table 8: Table 9: Table 10: Table 11: Table 12: Table 13: Summary of basic options (states) for tank car design safety features (acronyms for certain options in parentheses) (Barkan, 2008) Matrix of all possible combinations of risk reduction options considered for both insulated and non-insulated scenarios (and abbreviations used to denote them) [Barkan, 2008] Estimated CPR for non-insulated, non-pressure tank cars with different combinations of risk reduction options (Barkan, 2008) Estimated CPR for insulated, non-pressure tank cars with different combinations of risk reduction options (Barkan, 2008) Probability of release (CPR) from fleet given initial proportion of risk reduction options Probability of release (CPR) of fleet given 100% proportion of fleet being 5/8 inch tank cars Probability of release (CPR) of fleet given 100% of tank car fleet having Full Height Shield Protection Probability of release (CPR) of fleet given 100% of tank car fleet having Top Fittings Protection vi

8 Table 14: Table 15: Table 16: Table 17: Probability of release (CPR) of fleet given 100% of tank car fleet being Insulated Cars Average CPRs given 100% of fleet being made up of a particular risk reduction option Average CPRs given 100% of fleet being made up of a particular pair combination of risk reduction options Average CPRs given 100% of fleet being made up of a particular triad combination of risk reduction options vii

9 LIST OF FIGURES Figure 1: Figure 2: Figure 3: Components of a Non-jacketed, Non-pressurized Tank Car (Fritelli et al., 2014) Accident-Caused Releases in TIH Tank Cars (Jeong et al, 2009) Average weekly carloads of crude oil and petroleum products (EIA, 2014) Figure 4: U.S. Rail Carloads of Ethanol (AAR, 2013) Figure 5: U.S. Ethanol Production (AAR, 2013) Figure 6: Figure 7: Sequence of events leading to a hazardous release incident (Lui et al, 2014a) Procedure for assessing the distribution of the number of tank cars releasing per train derailment (Lui et al, 2014a) Figure 8: Schematic of Monte Carlo Simulation Model (Jeong, 2009) Figure 9: Overview of approach for evaluating likelihood of puncture (Sharma and Associates, 2014) Figure 10: Parent and child variables of a node Figure 11: Interpretation of presence or absence of links Figure 12: Bayesian Network modelled from CPR estimate data from Barkan (2008) Figure 13: Moral graph of figure Figure 14: Chordal graph of figure Figure 15: Chain of cliques with separator Figure 16: Bayesian Network of Total CPR model based on Barkan 2008 data viii

10 Figure 17: Undirected graph of figure Figure 18: Chordal Graph of figure Figure 19: Clique of chordal graph Figure 20: Figure 21: Figure 22: Bayesian Network given 100% proportion of fleet being 5/8 inch tank cars Bayesian Network given 100% of tank car fleet having Full Height Shield Protection Bayesian Network given 100% of tank car fleet having Top Fittings Protection Figure 23: Bayesian Network given 100% of tank car fleet being Insulated Cars ix

11 ABSTRACT Risk managers assessing hazardous materials release risk along various railroad routes and regions are tasked with evaluating the average likelihood of hazmat release from a derailed fleet of tank cars given varying proportions of tank car safety designs and operating conditions. These variations or changes may be as a result of retrofitting or phasing out of existing safety features (which have been deemed outmoded or unacceptable in the new safety climate), tank car fleet upgrade, construction of tank cars with new specifications, enhanced advanced braking rules or varying operating speeds. This thesis seeks to present Bayesian Networks (BNs) as a viable approach for modelling and supporting decision making in the fields of hazardous materials transportation risk and rail tank car safety. This approach estimates the average Conditional Probability of Release (CPR) of an existing or projected fleet of cars plying a given railroad route or region. CPR is one of the two primary components used in the analysis of hazardous materials release risk. This methodology can be used in assessing the reduction (or otherwise) of the average CPR of an existing or proposed fleet of tanks cars given a change in risk reduction option (tank car design safety feature or operating conditions). BNs allow for the evaluation of the effect of new or alternate risk reduction options (RRO) on the total x

12 network. They can also be used to evaluate the merits and demerits of the practice of grandfathering from a release probability point of view. Furthermore, Bayesian Networks can be used to easily rank the effect of various safety features and operating conditions given a CPR estimate dataset of all possible state combinations of the variables (risk reduction options) being considered. This allows researchers and decision makers to make decisions on which RRO to employ. As a result of interactive and flexible nature of BNs, these models can be integrated with other models to arrive at such a decision. The resulting average CPR value obtained from these models can be subsequently incorporated into the analysis of hazmat transportation risk. A CPR estimate dataset of possible combinations of four tank car design safety features was considered in the study. The features were tank thickness, insulation, head shield protection and top fittings protection. The aforementioned along with the total CPR made up the random variables of the Bayesian Network. The BN modelling was implemented using the commercially available HUGIN software. The average CPRs of the tank cars were computed given varying proportions of risk reduction options combinations. Sensitivity analysis was conducted to investigate the effect of various risk reduction options on the CPR of the fleet which were subsequently ranked. xi

13 Chapter 1 INTRODUCTION 1.1 General The safety of railroad tank cars has become imperative and of great national concern in light of recent train derailments that resulted in the release of hazardous materials and the explosion of tank cars. The subsequent evacuation and cleanup activities have been expensive and have drawn unwanted attention. However, there has been remarkable improvement in the crashworthiness of the tank car through coordinated efforts of industry, academia, and government in the areas of statistical analysis and design optimization, modelling, and impact-testing as well as in the development of tank car design standards. The crashworthiness of the tank car given various tank car designs and operational conditions can be evaluated by its Conditional Probability of Release (CPR). Researchers and railroad officials in the tank car safety industry are constantly faced with decisions to be made in relation to tank car standards or specifications, individual tank car safety features and operating conditions. These include decisions to be made on retrofitting or phasing out of safety features which are deemed outmoded or unacceptable in the new safety climate as well as evaluation of new or alternate features that may be incorporated into future specifications. Decision makers rely on CPR tank car estimates as an industry index used in the evaluation of accident performance or puncture resistance of tank cars during derailments (Barkan, 2008, Saat and Barkan, 2011). The CPR of a derailed tank car is dependent on its design 1

14 features and derailment speed (Kawprasert and Barkan, 2010, Lui et al, 2014a), and can be affected by commodity type (Glickmann et al, 2005). CPR is one of the two principal components employed in the evaluation of hazardous materials transportation risk (Kawprasert and Barkan, 2010). Hazmat Release risk is defined as the product of the frequency of the undesirable (release) incident and its corresponding consequence (Glickman and Erkut, 2007, Kawprasert and Barkan, 2010). Bayesian networks can be used in tandem with existing models in evaluating the average likelihood of release of hazmat materials from a derailed tank car given changes or variations in risk reduction options. This allows decision makers to easily quantify the average CPR of given tank car fleets that ply a given railroad route. The resulting estimate is subsequently incorporated into the analysis of hazmat transportation risk. Furthermore, the reduction (or otherwise) of the average probability release as a result of these changes or variations can be assessed. Thus the effect of new or alternate tank cars, features or operating conditions on the total network can be compared. It can also be used to evaluate the merits or demerits of the practice of grandfathering from a release probability point of view. Glickman et al. (2007), in their assessment of tank car risks along various railroad routes with the primary aim of the analyzing the tradeoff between cost and risk impacts, assumed a CPR of 25% for all routes in their analysis. The authors defined CPR as the percentage of tank cars damaged or derailed in train accidents that experience a major release. It can be assumed that they selected this value based on prior knowledge. Thus, one can safely assumed that their analysis did not take into account the heterogeneous nature of tank car traffic and their operating conditions along various routes. Bayesian Networks can help researchers and risk managers more 2

15 accurately estimate the CPR of tank car traffic along certain routes given the nature of the fleets and their operating conditions. In the case of several types of cars being used but the effect of the varying design characteristics not being of interest, Kawprasert and Barkan (2008) suggested the computation of the aggregated CPR using the weighted mean equation of the different car types conditional probabilities. Bayesian Networks on the other hand, can be used in not only easily finding the weighted mean or average CPR of the fleet of tank cars but can also be used to easily observe the effect of the varying design characteristics of interest. Furthermore, Bayesian Networks can be used to easily rank the effect of various safety features and operating conditions (on the average CPR) given a CPR estimate matrix of all possible state combinations of the variables (risk reduction options) being considered. This allows researchers and decision makers decide on which risk reduction options to employ. This may not necessarily be the optimal combination of risk reduction options based on minimal CPR; other factors such as weight, capacity, cost or incremental safety benefit may be considered in selecting the most appropriate set of risk reduction options to be utilized. As a result of BN s interactivity and flexibility, it allows this model to be integrated with other models to arrive at such a decision. Logistic regression models have been widely used in industry for CPR estimation. However like most models they have their disadvantages and limitations. These include their inability to handle both continuous and discrete variables, inability to incorporate subjective and incomplete information as well as the difficulty for 3

16 sensitivity analysis to be conducted. Bayesian Networks addresses the aforementioned limitations in the Logistic Regression models. Given access to extensive accident data, Bayesian Networks can also be utilized by decision makers as a viable alternative to existing models in the evaluation of CPR. The formulation of Bayesian Networks models given the individual probability distribution of the various variables affecting the Conditional Probability of Release obtained from accident data are beyond the scope of this thesis. 1.2 Statement of Problem Railroad Tank Car safety has been an issue of great national concern in recent years. It has become imperative in light of recent high profile rail accidents involving hazardous materials release along with the surge in crude oil production and its transportation by rail tank cars in North America. Hazmat release may result in safety implications for all stakeholders as well as financial cost through the subsequent property damage, evacuations, service disruptions, environment cleanup and litigation that may arise (Barkan et al, 2008). One of the most recent incidents was on the 16th of February, 2015 in the Fayette County where the Governor of the State of West Virginia declared a state of emergency due to the derailment and subsequent explosion of a freight train carrying crude oil. This accident resulted in the evacuation of over a thousand residents and the shutdown of a local water treatment plant. The ramifications of such events and the level of risk involved in rail tank car operations underscore the importance of advanced research in this area. 4

17 Risk managers assessing hazmat release risk along various railroad routes and regions are faced with evaluating the average likelihood of hazmat release of a fleet of tank cars given a derailment. This thesis seeks to demonstrate the use of Bayesian Networks with existing models in evaluating the average CPR of a proposed or existing fleet of tank cars that ply a given railroad route given changes or variations in risk reduction options, which can be subsequently incorporated into the analysis of hazmat transportation risk of various railroad routes and regions. Furthermore, they can be used in making decisions in relation to tank car standards or specifications, individual tank car safety features as well as operating conditions. These include decisions to be made on retrofitting or phasing out of safety features which are deemed outmoded or unacceptable in the new safety climate. Other decisions include the evaluation of effects of tank car fleet upgrade or new or alternate features which may be incorporated into future specifications. 1.3 Objective of Research The main objective of this research is to develop new probabilistic graphical models for estimating the Conditional Probability of Release (CPR) of hazardous materials from tank cars given a train derailment. These models will be used in evaluating the average likelihood of hazmat release given varying proportions or changes in risk reduction options. This will be achieved through the following subobjectives: 1. To identify the various variables or risk reduction options that form input into the CPR analysis. 2. To formulate Bayesian Network CPR model based on the aforementioned variables being considered. 5

18 3. To develop a model that can incorporate subjective and incomplete information. 4. To develop algorithms though which sensitivity analysis can be conducted. 5. To investigate the effect of various risk reduction options on the CPR of a fleet of tank cars. 1.4 Research Framework The research framework of this thesis entails: A. State of the art of existing CPR models B. Identifying the various variables or risk reduction options C. Formulating the CPR graphical model D. Analyzing the CPR graphical model 1.5 Organization of the Thesis The thesis has six main chapters. Chapter 1 is the introductory chapter and contains a brief introduction to the topic, the statement of the problem, objectives of this research, the research framework and the structure of the thesis. Chapters 2 to 4 contains the background and literature review. Chapter 2 outlines the history of the tank car, describes the various tank car specifications and improvements being made to tank car safety design. This chapter also describes train accidents involving tank car release as well as transportation of flammable liquids such as crude oil and ethanol by rail. Chapter 3 describes the Conditional Probability of Release and the various CPR models used such as Logistic Regression models and Monte Carlo methods. It also identifies the limitations of the CPR approach and 6

19 describes other models used in industry. Chapter 4 describes the concept of Bayesian Networks. Chapter 5 describes the dataset and BN model used in the analysis, and summarizes the analysis made. It also contains a discussion of the results of the analysis. Chapter 6 subsequently outlines the conclusion and recommendations made. 7

20 Chapter 2 BACKGROUND AND LITERATURE REVIEW 2.1 History of the Tank Car Steps to enhance the safety and efficiency of tank cars started well over a century ago. Tank cars were first constructed to haul crude oil from western Pennsylvania to mainly East Coast refineries and markets. The Master Car Builders Association (MCBA) have been involved in the development and maintenance of several tank car constituent standards since However, it was not until 1903 that tank car design specifications were introduced as a result of the collision and subsequent inferno in Sharon, Pennsylvania in 1902 involving several tank cars on the Pennsylvania Railroad (Barkan, 2008). This initial set of recommendations by the Association s Tank Car Committee (TCC) comprising of representatives from several railroads and the leading tank car owner, Union Tank Line Company became industry standards in 1910 following several amendments that were subsequently adopted by the American Railway Association, the predecessor of Association of American Railroads (AAR). In 1912, two extra risk-based standards were introduced, which required the construction of stronger tank cars for the special haulage of Casinghead gasoline and chlorine. The AAR Tank Car Committee is currently mandated to assess and alter specifications (Barkan et al 2013, Barkan 2008). There have been technological improvements to the tank car notably material and tank-fabrication-related advancement with fusion-welding technology replacing 8

21 forge-welding technology and riveted tank construction. Steel making advancements have led to increased puncture resistance, fracture toughness, weldability and formability of manufactured steel; all are key attributes of steel needed for improved tank car design and construction. Other materials other than carbon steel that have been used in the construction of tank cars include stainless steel, aluminum and nickel (Barkan, 2008). 2.2 Tank Car Standards The oversight of federal regulations pertaining to tank car standards were initially placed in the hands of the now-defunct Interstate Commerce Commission (ICC) in The ICC on 1 st of July, 1925 issued new tank car specifications and two years later the ICC required design acceptance by the TCC. Industry standards became Department of Transportation (DOT) regulations on the 1 st of April, 1967 (Barkan et al, 2013; Fronczak, 2014). Tank Car specifications vary greatly with the robustness of the cars depending on the hazard levels of the products they are intended to transport (Barkan, 2008) Legacy DOT 111 Tank Cars DOT 111 tank cars are the most common type of tank cars in North America, and DOT-111 specification is the minimum tank car specification intended for hazardous liquids that pose low to moderate hazard. They are non-pressurized tank cars, and the focus of their design has been on issues such as puncture resistance and top fitting protection. They are about 60 feet in length, 11 feet in width and 16 feet in height. They have a tare weight (weight when empty) of 80,000 pounds and a gross weight (weight when full) of 286,000 pounds. The capacity of the tanks range from 9

22 13,000 to 33,000 gallons (49, ,000 liters) (Fritelli et al, 2014; USDOT, 2015b; Barkan, 2008). Unlike other types of railcars, tank cars not only have transportation functions but important storage functions as well. Tank Cars are relatively lowly utilized with an annual average of 9 trips and an average trip cycle of 6 weeks (Barkan, 2008). About 69% of tank cars in operation are DOT-111s, transporting a wide spectrum of hazmat commodities. The total number of tank cars in North America are 334,869 with 228,036 being DOT-111s of which 97,744 of them transport flammable liquids (42,611 for crude oil, 29,785 for ethanol and 39,813 for other flammable liquids) (Stancil, 2014; Fronzcak, 2014). Previous Investigations into DOT-111 Tank Car Design include a Safety Study in 1991, and an investigation into train derailment incidents with significant tank car release including Superior, Wisconsin (1992), Tamaroa, Illinois (2003) and New Brighton, Pennsylvania (2006) (Stancil, 2014). These tanks have been found to almost always breach given a train accident resulting in car-to-car impacts or pileups. This was indicated in the (June 19, 2009) Cherry Valley, IL NTSB published findings. DOT -111 tank cars are known to be only safe from collisions for speeds up to 9 miles per hour (USDOT, 2015b). The expected service life of these tank cars estimated to be a maximum of 50 years with an economic life of about 30 to 40 years (Fronczak, 2014). In light of this, the AAR and DOT have in the past allowed for the permissive, continued use of tank cars conforming to former regulatory standards which is known as grandfathering. The continued use of earlier technology was allowed with the assumption it does not reduce safety unless it presented an intolerable risk. However, there has been 10

23 increased opposition to this controversial practice in light of several accidents involving toxic inhalation hazard (TIH) materials release. A notable example which led to great concern was the continued use of pressure-specification tank cars made from non-normalized steel despite its ban from new pressure car construction in The AAR eventually called for the expedited retirement of the existing fleet for TIHs in favor of more robust designs (Barkan 2008) CPC 1232 Tank Cars The AAR during the aftermath of these derailments (Stancil, 2012) subsequently issued Casualty Prevention Circular (CPC) 1232, an interchange standard which was voluntarily adopted by industry. This took effect on October 1, The CPC-1232 outlined industry requirements for DOT-111 tank designated for use in crude oil and ethanol service with the aim of enhancing the crashworthiness of these tank cars. These improvements include thicker head protection (minimum ½ inch half-height head shields on both ends), top fitting protection, pressure relief valves with greater flow capacity, and either a thicker tank shell and head (minimum ½ inch and 7 / 16 thick for unjacketed and jacketed cars respectively) or a jacket [minimum ½ inch and 7 / 16 thick for unjacketed and jacketed cars respectively]. However there were no requirements for high capacity pressure relief valves or bottom handles (Fronczak 2014, USDOT 2015a, Lui et al, 2014a). Despite these enhancements and the more robust protection, these CPC-1232 cars have been involved in derailments that resulted in hazmat release. Most recent examples of such accidents involving CPC-1232s include Mt. Carbon, WV, Dubuque, IA, and Galena, IL all occurring in 2015 (USDOT, 2015a). 11

24 2.2.3 DOT 117 Tank Cars In response to such incidents, the PHMSA published a notice of proposed rulemaking (NPRM) on August 1, The PHSMA, in conjunction with the FRA subsequently published an advanced notice of proposed rulemaking (ANPRM). This ANPRM aimed to seek comments for potential revisions to proposed amendments to existing hazard material regulations (HMR) relating to the transportation of flammable liquids by rail in large quantities. The final rule titled Hazardous Materials: Enhanced Tank Car Standards and Operational Controls for High-Hazard Flammable Trains addressed comments on the NPRM made by industry and the public and made amendments to the HMR (USDOT, 2015a). On May 1, 2015, the USDOT, through the FRA and PHMSA, in conjunction with Transport Canada issued new stringent and enhanced tank car standards and operational controls for High-Hazard Flammable Trains (HHFTs). The tank car specifications for the new proposed non-pressurized tank car known as DOT-117 in the United States and TC-117 (formerly TC-140 prior to final rulemaking) in Canada adopted the most demanding of the technical requirements first offered for comment in the notice of rulemaking (USDOT, 2015b; David Thomas, 2015b). Three new alternative tank car options were proposed in the NPRM: PHMSA and FRA Design Tank Car (Option 1), AAR 2014 Tank Car (Option 2) and Enhanced CPC-1232 Tank Car (Option 3). A modified version of Option 2 was adopted by the PHMSA and FRA for new construction of new tank cars to be used in HHFTs. Option 3 was adopted for the retrofitting of existing tank cars to be used in HHFTs (USDOT, 2015a). The adopted DOT-117 safety enhancements include full-height ½ inch thick head shields, minimum 9 / 16 inch TC-128 Grade B normalized steel, thermal protection, 12

25 minimum 11-gauge jacket, top fittings rollover protection, sturdier re-closeable pressure relief valves and enhanced bottom outlet handle design to prevent unintended actuation during a train accident (USDOT, 2015b). The higher tensile strength TC 128 Grade B steel recommended for use in DOT-117 cars are already in use in other higher gross rail load (GRL) cars. This steel has been found to have a 15.7% higher tensile strength than A-516 which is commonly used in legacy DOT-111 tank cars (Barkan, 2008). The aforementioned regulatory bodies in the issue of this final rulemaking defined HHFTs as a continuous block of 20 or more tank cars loaded with a flammable liquid or 35 or more tank cars loaded with a flammable liquid dispersed through a train. High-Hazard Flammable Unit Trains (HHFUTs) are also defined as trains comprised of 70 or more loaded tank cars containing Class 3 flammable liquids traveling at speeds greater than 30 mph (USDOT, 2015b). After October 1, 2015, all newly-constructed cars intended for use in HHFTs are to meet the new tank car design standards (or performance criteria). On the other hand, existing cars are mandated to undergo retrofitting conforming to retrofit design or performance standard for use in an HHFT prescribed by USDOT, or else face the risk of being phased out. In addition, the coalition of regulatory bodies have drawn up retrofit deadlines which differ for various packing group (PG) of flammable liquids as well as various types of CPC-1232 and DOT-111 cars (USDOT, 2015b). The new ruling limits HHFTs to operating speed restrictions of 50-mph with further restrictions of 40-mph in high-threat urban areas placed on HHFTs containing any tank cars which fail to meet the new tank car specifications. Furthermore, enhanced braking rules were adopted which mandate HHFTs to have in place a 13

26 distributive power (DP) braking system or a functioning two-way end-of-train (EOT) device with electronic controlled pneumatic (ECP) braking system required in HHFUTs. ECP braking systems are required by January 1, 2011 in HHFUTs transporting at least one PG I flammable liquid with all other HHFUTs required to meet a May 1, 2023 deadline. The new regulations require railroads operating HHFTs to map out suitable routes based on results from routing analysis which take into consideration a minimum of 27 safety and security factors to make an assessment. Railroads are further required to grant information access of such decisions to state and/or regional fusion centers as well as state, local and tribal officials who request for it. Results of well-documented testing and sampling programs for all unrefined petroleum-based products (such as crude oil) are to be readily available upon request by USDOT personnel (USDOT, 2015b). These new DOT-117 tank cars being proposed will lessen the cleanup exercise for most low-speed mishaps (Thomas, 2015a). They are expected to reduce the likelihood of lading release given impact from a derailment by about 85% in comparison with the current non-jacketed DOT-111 cars (Barkan et al., 2015); however, there have been concerns raised about the effectiveness of the new DOT-117 specifications. Option 1, which had the most robust design and was rejected in favor of option 2 as a result of additional cost without any substantial increase in safety, was found to have safe speed for collisions up to only 12.3 miles per hour for the shell of the tank (USDOT, 2015a). The AAR-proposed tank car (option 2) and the thinner CPC-1232 Plus (option 3) which were eventually selected as DOT-117 specifications for new tank cars and retrofitted cars respectively were found to suffer simulated 14

27 couple punctures at mere 18.4 and 17.8 mph collision speeds. Despite being much higher than that of an unjacketed DOT-111 (8.6mph), either puncture threshold falls well below the speeds of the three most explosive Bakken oil train derailments: Lac- Mégantic (65mph), Casselton (42mph) and Aliceville (39mph). The DOT-117 will not prevent the fireballs of the magnitude of the aforementioned derailments (Thomas, 2015a). Currently affected tank cars that do not meet the prescribed DOT-117 specifications will have to be phased out or retrofitted by stipulated timelines. The timeline for retrofit of affected tank cars for use in North American HHFTs is shown in table 1 (USDOT, 2015a). 15

28 Table 1: Timeline for the Retrofit of Affected Tank Cars for Use in North American HHFTs (USDOT, 2015a) Timeline for the Retrofit of Affected Tank Cars for Use in North American HHFTs Tank Car Type / Service US Retrofit Deadline Tank Car Type / Service TC Retrofit Deadline Non Jacketed DOT- 111 tank cars in PG I service (January 1, 2017) January 1, 2018 Non Jacketed DOT-111 tank cars in Crude Oil service May 1, 2017 Jacketed DOT-111 tank cars in PG I Non Jacketed CPC tank cars in PG I service Non Jacketed DOT- 111 tank cars in PG II service Jacketed DOT-111 tank cars in PG II service Non Jacketed CPC tank cars in PG II service Jacketed CPC-1232 tank cars in PG I and PG II service and all remaining tank cars carrying PG III materials in an HHFT (pressure relief valve and valve handles). March 1, 2018 April 1, 2020 May 1, 2023 May 1, 2023 July 1, 2023 May 1, 2025 Jacketed DOT-111 tank cars in Crude Oil service Non Jacketed CPC tank cars in Crude Oil service Non Jacketed DOT-111 tank cars in Ethanol service Jacketed DOT-111 tank cars in Ethanol service Non Jacketed CPC tank cars in Ethanol service Jacketed CPC-1232 tank cars in in Crude and Ethanol service and all remaining tank cars carrying PG III materials in an HHFT (pressure relief valve and valve handles). March 1, 2018 April 1, 2020 May 1, 2023 May 1, 2023 July 1, 2023 May 1,

29 2.3 Tank Car Design Improvement Despite these DOT-117 limitations, there have been ongoing efforts to improve the design and safety performance of the tank car. Efforts by the Next-Generation rail tank car coalition (NGRTC) has resulted in substantial research into tank car crashworthiness which includes material and full-scale shell impact testing, computer modeling of train derailments, tank car response and tank car design optimization modelling. This coalition initially comprised of Dow Chemical, Union Pacific Railroad and Union Tank Car Company in 2006 and later incorporated other stakeholders such as USDOT, FRA, Transport Canada, academia and other industry partners (Barkan et al, 2013; Jeong, 2009). One tank car enhancement concept under development is that of engineered sandwich structures. These structures are pressurized tanks enclosed in a separate exterior carbody made of sandwich panels designed to withstand all in-service loads thus removing the tank from the load path during normal operations. Sandwich structures are stronger than solid plates of the same weight. Tank car protection is achieved through load-blunting of impactors and absorption of collision energy (Jeong et al., 2009). One coup of the NGRTC is the development and validation of detailed finite element models of tank car equipment which can accurately predict the puncture prediction capacity or resistance under varying impact conditions. This allowed for the quick estimation and comparison of varying tank car designs, and impact condition alternatives as well as the evaluation of a variety of factors associated with impact safety. The developed analytical tank impact algorithms as well as the computed relative puncture energies for the various tank design alternatives are valuable in making safety decisions (FRA, 2013). 17

30 A larger coalition called the Advanced Tank Car Collaborative Research Program (ATCCRP) was formed in 2009 to continue the work on the NGRTC. They developed two initial projects aimed at improving tank car design performance assessment. One project was set up to identify the most appropriate failure criteria in modeling the performance of tank steels, as well as the material properties to support accurate use of such criteria. The other project was set up to simulate a variety of tank head and shell impact scenarios to estimate the energy absorption of each tank car design (Barkan et al, 2013). The systematic approach to improved tank car design development entails three steps: 1) Defining the collision conditions of concern 2) Evaluation of the structural behavior of conventional tank car design under these conditions 3) Development and evaluation of alternate performance improvement designs (Jeong et al, 2009). The three main approaches of assessing tank car damage resistance are mainly through testing, modelling and statistical analysis of empirical data. Statistical Analysis has many advantages over testing and modelling in that an entire spectrum of actual accident scenarios can be accounted for. The same cannot be said for testing or modelling of accident scenarios since only one idealized scenario per test or run respectively (Treichel, 2014). Furthermore, statistical analysis has a huge advantage in terms of cost per analysis which turns out to be low once model is developed whereas testing cost per analysis is very high and modelling cost per analysis is moderate once model is 18

31 developed. Additionally, statistical analysis has the ability to quantify expected releases in operation however this is not currently possible using testing and modelling approaches. The downside of using the statistical analysis of empirical data approach is its limited ability to analyze new design elements not already in fleet which is well catered when using the testing and modelling approaches to assessing damage resistance (Treichel, 2014). Modelling Software such as LS-DYNA and ABAQUS are used to create simulations of impact scenarios using dimensions and measurements of the tank car. Crashes are performed using the Transportation Technology Center (TTC) Crash Wall for repeatable impact testing of the tank cars (Gonzalez, 2014). 2.4 Train Accidents Involving Tank Car Release Ever since the 1902 Sharon, Pennsylvania derailment and subsequent tank car release which led to the establishment of tank car specifications (Barkan, 2008), train accidents involving hazmat release from tank cars have been in the public limelight and have been a safety concern for both industry and government. Over 99% of hazmat shipments are transported safely and incident-free (Lui et al, 2014a). Furthermore, the annual number of accidents with at least one car releasing hazmat contents has drastically reduced over the last quarter of a century (FRA, 2013); however, despite the fatalities and injuries due to hazmat rail accidents being far lower than that of traffic accidents, there has been extreme public consternation due to the involuntary nature of risk and the high consequence potential of these incidents (Glickman et al, 2007). The heavy investment in railroad infrastructure since economic deregulation had led to drastic decline in the railroad accident rates and hazmat release rates in the two decades prior to the early 2000s. The upgrade in infrastructure quality led to faster 19

32 train speeds, enhanced reliability and less derailments (Barkan, 2008). However, various accidents involving fatal hazmat releases in the mid-2000s rekindled an interest in tank car safety despite a more than 90% reduction in the rate of hazmat release from 1980 to 2012 (Barkan et al, 2013). This succession of incidents have highlighted deficiencies in our engineering safety systems resulting in the safe movement of flammable liquids such as crude oil and ethanol becoming politically challenged (Johnson, 2014). Examples of such incidents in the last quarter of a century are listed in table 2 (Jeong, 2009; Stancil, 2012; Stephens, 2014; El-Sibaie, 2014; US DOT, 2015b, FRA 2015; TSB, 2015). 20

33 Table 2: List of recent train derailments in North America occurring over the last quarter of a century (Jeong, 2009; Stancil, 2012; Stephens, 2014; El- Sibaie, 2014; US DOT, 2015b; FRA 2015; TSB, 2015). Town/City State Country Day Month Year Superior Wisconsin United States 30 June 1992 Minot North Dakota United States 18 January 2002 Tamaroa Illinois United States 9 February 2003 Macdona Texas United States 28 June 2004 Graniteville South Carolina United States 6 January 2005 Anding Mississippi United States 10 July 2005 Texarkana Arkansas United States 15 October 2005 New Brighton Pennsylvania United States 20 October 2006 Shepherdsville Kentucky United States 16 January 2007 Oneida New York United States 12 March 2007 Painesville Ohio United States 10 October 2007 Luther Oklahoma United States 22 August 2008 Rockford Illinois United States 20 June 2009 Arcadia Ohio United States 6 February 2011 Tisilkwa Illinois United States 7 October 2011 Columbus Ohio United States 11 July 2012 Plevna Montana United States 5 August 2012 Lac-Mégantic Quebec Canada 6 July 2013 Aliceville Alabama United States 8 November 2013 Casselton North Dakota United States 30 December 2013 Lynchburg Virginia United States 30 April 2014 Vandergrift Pennsylvania United States 13 February 2014 New Augusta Mississippi United States 31 January 2014 Plaster Rock New Brunswick Canada 7 January 2014 Mount Carbon West Virginia United States 16 February 2015 Galena Illinois United States 5 March 2015 Dubuque Iowa United States 4 February 2015 Timmins Ontario Canada 15 February 2015 Gogama Ontario Canada 7 March

34 Hazmat release incidents are usually of very high consequence since rail hazmat haulage are mostly done in bulk shipments (Glickman et al, 2007). Furthermore, the low probability-high consequence nature of these incidents results in significant attention both from the public and media (Glickman and Erkut, 2007). Hazmat release due to train accidents may result in safety implications for all stakeholders. These accidents may also result in financial cost through the subsequent property damage, evacuations, service disruptions, environment cleanup and litigation that may arise. These accidents tend to cost millions or tens of millions of dollars (Barkan, 2008). The December 30, 2013 Casselton derailment as well as the July 6, 2013 Lac-Mégantic derailment and their subsequent explosions and inferno led to the mass evacuation of the affected areas with 2000 and 1400 residents moved to safety respectively. The latter resulted in 47 fatalities and massive destruction to the town which led to the introduction of legislation by Congress requiring railroads to have at least 2 crew members aboard all trains (Frittelli et al., 2014). Table 3 lists some of these incidents along with information such as location, the number of cars derailed, derailment speed, lading loss quantity and cause of derailment (Stephens, 2014; El- Sibaie, 2014, FRA 2015). 22

35 Table 3: List of recent train derailments along with key information on each incident (Stephens 2014, El Sibaie 2014, FRA 2015) # Cars derailed Speed at derailment Unit train Product Loss (gal) Incident Year New Brighton, PA Yes 485,278 Rail Painesville, OH No 76,153 Rail Luther, OK Yes 243,000 Cross level Rockford, IL No 232,963 Washout/Rail Arcadia, OH Yes 834,840 Rail Tiskilwa, IL No 143,534 Broken Rail Columbus, OH No 53,347 Broken Rail Cause of Derailment Irregular Track Alignment Plevna, MT No 245,336 Lac-Mégantic, QC (Canada) Yes 1,580,000 Securement Aliceville, AL Yes 700,000 Rail Casselton, ND Yes 400,000 Broken Axle on an Adjacent Train Wide gage as a result of broken Spike Vandergrift, PA Yes 10,000 New Augusta, MS Yes 90,000 Broken Rail Plaster Rock, NB (Canada) No 30,000 Broken Rail/Wheel The major cause of freight train derailments based on Class I railroad mainline track accident data was broken rail and track welds (670). This accounted for almost a quarter (23%) of derailed cars with track geometry defects (300) a distant second. Furthermore, the former was found to have a higher accident severity with an average of 13 derailed cars recorded compared to 8.6 for other causes. In a separate investigation, track problems were found to be the most important causes of 23

36 derailment in the same time frame with train equipment problems next in line (Frittelli et al, 2014). Ways of reducing derailment severity and their consequences include improving tank car design and reducing the kinetic energy in a derailment. Kinetic energy reduction can be achieved by reducing the operating speeds of trains or implementing advanced braking systems (Gonzalez et al, 2014). There are four primary sources of lading loss: tank head, tank shell, top fittings and bottom fittings (Jeong, 2009; Barkan, 2008) and are illustrated in figure 1 (Fritelli et al., 2014). There are two main types of release-causing damage to the car: damage or puncture to the tank head or shell (also known as tank sources or causes) and damage to fittings and appurtenances such as top and bottom fittings (also known as non-tank sources ). It is important to differentiate the two causes due to the differing nature and effects of damage as well as their related risk and mitigating strategies employed (Barkan et al, 2007). Average tank head and shell average lading losses have been found to be much greater than that of top and bottom fittings (Saat and Barkan, 2011). 24

37 25 Figure 1: Components of a Non-jacketed, Non-pressurized Tank Car (Fritelli et al., 2014)

38 Analysis of the RSI-AAR Project Accident database comprising 252 damaged tank cars that released toxic inhalation materials (TIH) in 176 accidents between 1965 and 2005 showed that 81% of the volume of lading lost was due to tank head (36%) and shell (45%) failure. This was in spite of these failures accounting for about half (about 23% and 27% respectively) of all releases. On the other hand, failures due to valves and fittings accounted for 27% of total accident-caused releases but a mere 1% of total gallons lost. Thus the frequency of head and shell releases are slightly lower than the total of all other causes, but they have a much higher consequence (Jeong et al, 2009). This is illustrated in Figure 2 (Jeong et al, 2009). One explanation for such a disparity is that it is easy to halt a fittings leakage; however, a tank puncture in the shell or head is usually much bigger and harder to seal (Saat and Barkan, 2011). Figure 2: Accident-Caused Releases in TIH Tank Cars (Jeong et al, 2009). 26

39 2.5 Transportation of Crude Oil by Rail The recent surge in North American crude oil supply has resulted in the continent supplying about two-thirds of domestic demand overtaking imports. Some areas in the United States are on course to record excess supplies of oil and refined products. This growth is largely due to increasing production in the Canadian oil sands in addition to that of shale oil production from the Eagle Ford and Permian Basins in Texas as well as the Bakken Fields of North Dakota and Montana (Frittelli et al, 2014). Crude oil and ethanol shipments account for about 68% of flammable liquids hauled by rail (USDOT 2015). Crude oil shipments by freight railroads have increased from 9,500 carloads in 2008 to 434,000 carloads or approximately 300 million barrels in 2013, which translates to a 4570% increase in a spate of 5 years with crude imports from neighboring Canada increasing 20-fold since 2011 (Fritelli et al, 2014). United States northern neighbor has become its chief foreign supplier as a result of the oil sands production boom. In 2013, rail crude oil shipments exceeded the 400,000 carloads mark (USDOT, 2015) with 2.72 billion barrels domestically produced and an additional 2.82 billion barrels imported (Fritelli et al, 2014). 27

40 Figure 3: Average weekly carloads of crude oil and petroleum products (EIA, 2014). Oil train reform continues to be focused on the tank car despite the fact that the technology to make lading safer for transportation is cheap and widely available. The technology has already been employed in Texas where light crude is treated to remove explosive gases before loading. This narrowed focus has allowed oil producers to avoid any responsibility shifting all the financial and operation burden of reform to the shippers, fleet owners and the railroads (Thomas, 2015a). 2.6 Transportation of Ethanol by Rail Ethanol makes up 26% of total Hazmat shipments and 1.1% of total railroad shipments (RFA, 2013). Domestic ethanol production is mainly concentrated in corn growing areas, particularly Midwestern states such as Iowa, South Dakota, Nebraska and Illinois, and transported to major markets on the East Coast, Texas and California 28

41 by rail which makes up about 70% of ethanol shipments. However, most of the corn required for ethanol production is transported to plants by truck with few plants receiving the raw material by rail (AAR, 2013). Over the years, Ethanol have been transported in standard DOT 111A railcars with 85% of the current fleet less than 7 years (RFA, 2013). Most ethanol rail shipments are transported in 30,000-gallon tank cars with most of these cars owned by shippers or leasing companies and not the railroads (AAR, 2013). Similar to crude oil, ethanol production has significantly increased over the last 10 years (USDOT, 2015). Production has risen from 3.4 billion gallons in 2004 to about 13.3 billion gallons in 2013, which translates to a 291% increase (figure 5) peaking at 13.9 billion gallons in 2011 (AAR, 2013). Consequently, this has generated growth in ethanol transportation by rail (USDOT, 2015) with rail shipments rising by more than 500% from 51,000 carloads in 2003 to over 306,000 carloads in 2012 (Figure 4). Furthermore, ethanol rose from 0.2% of total rail carloads in 2003 to 1.0% in 2012, 0.3% of rail tonnage in 2003 to 1.5% in 2012 and 0.4% of rail ton-miles in 2003 to 2.0% in 2012 (AAR, 2013). 29

42 Figure 4: U.S. Rail Carloads of Ethanol (AAR, 2013) Figure 5: U.S. Ethanol Production (AAR, 2013) 30