Recommended Guidelines for Curb and Curb-Barrier Installations (2005)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

31 CHAPTER 4 RESEARCH APPROACH INTRODUCTION This chapter discusses the methods of analysis that were used in this study, including crash and geometric data analy- ses, computer simulation, and full-scale crash testing. Existing crash and geometric databases were examined to determine if they could be used to characterize the extent and severity of safety problems associated with curb and curb– barrier combinations. The crash databases were also reviewed to determine if they could provide information regarding the nature of impacts involving curbs (e.g., impact speed, angle of impact) in order to develop input for full-scale crash test- ing and computer simulation studies. Where validated computer models can be developed, com- puter simulation methods are the most versatile approach for investigating a wide range of possible impact scenarios (e.g., vehicle type, curb type, impact condition). Computer simula- tion can also be very useful for determining the precise effects that vehicle-curb interactions have on the stability of various vehicle types and the effects that curbs placed in com- bination with roadside safety barriers have on the perfor- mance of the barriers. Vehicle dynamics programs and Finite Element Analysis (FEA) are two such methods that were con- sidered for use in this study. Vehicle dynamics programs have been used extensively in previous curb-safety-related studies, as indicated in the literature review (Chapter 2). FEA has been used in several studies involving vehicle impact with roadside safety hardware and has proven to be very effective. To the knowledge of the authors, however, FEA has not been used in any study involving curbs or curb–barrier combina- tions and, therefore, was not discussed in the literature review section of this report. Since FEA was an important analysis tool in this research, the effectiveness of the method applied in the study of roadside barrier crashworthiness is discussed in this chapter. A summary of previous studies using FEA to study vehicle impact with roadside safety barriers is presented and discussed later in this chapter. Full-scale crash testing was another method used in this research. The advantage of full-scale crash tests is that they are actual physical impact events in which there is little ambiguity about the results. The disadvantage is that they are costly, and it is seldom feasible to perform very many tests. The testing results, therefore, usually do not address a very wide range of conditions. A full-scale testing program was used in this project to verify and confirm hypotheses devel- oped from the computer simulation study, as well as to vali- date and strengthen the conclusions of this research. ANALYSES OF CURB-RELATED SAFETY ISSUES USING CRASH AND INVENTORY DATA Introduction Since the inception of this study, an overall goal has been to use existing databases containing information on crashes, roadway inventory, and traffic to better characterize safety problems associated with curb and curb–barrier combinations on higher-speed roadways. Such information was used directly in the development of the design guidelines since it can pro- vide real-world insight into the magnitude of the problem on various roadway types, the nature of the problem (e.g., how curb impacts are similar or dissimilar to other run-off-road collisions), and factors that might be influenced to reduce curb impact severity (e.g., to prevent rollover after a curb impact). In addition to this primary goal of input into design guidelines, a secondary but related goal of the crash-data analyses was to provide leads for the crash testing and simu- lation efforts that were later conducted. The crash-data analyses took place in two phases. Phase I involved a detailed examination of existing databases to determine which ones might be suitable for use. Based on preliminary examination of data and discussions with the proj- ect panel, a final set of crash-data analyses were defined. These analyses were then carried out in Phase II, using the selected databases. Examination of Databases As detailed in an interim report, the national databases of interest included the following: • Fatality Analysis Reporting System (FARS) • National Automotive Sampling System—General Esti- mates System • National Automotive Sampling System—Crashworthi- ness Data System • FHWA’s Highway Safety Information System (HSIS)

Preliminary examination of each of these databases was under- taken to determine whether it would be useful in the overall safety analysis and what types of analyses might be possible with it. Fatality Analysis Reporting System (FARS ) This database is an annual census of all police-reported fatal- ities in the United States, with data coded and cleaned by a FARS coder in each state. Data from the 1994–99 period were used in this effort. FARS contains data on the presence of a “Curb” as both First Harmful Event (FHE) and Most Harmful Event (MHE), and data on “Rollover” separate from the event codes. FARS does not contain data on the full “Sequence of Events” in a crash (e.g., curb strike, then guardrail impact, then overturn). FARS does not contain any information on curb design parameters. Finally, since it is based on fatal crashes, FARS data could not be used to examine differences in injury severities with and without curbs; only the fatal crash failures are present. National Automotive Sampling System— General Estimates System (NASS-GES) The GES was established by NHTSA to allow national estimates of safety issues. It contains annual files for 1988 and later. Data from the 1995-99 period were used in this analysis. GES is based on an annual random sample of approx- imately 50,000 police crash reports of all severities (ranging from no-injury to fatality) pulled each year from 60 areas (400 police jurisdictions) across the nation. All cases are manually coded to approximately 90 common data elements. The coding is based on a review of the computerized codes on the original form and the narrative and sketch. GES assigns a weight for each case that allows one to develop national estimates; severity is the predominant weighting variable. As will be seen later, both weighted and unweighted data were used, depending on the nature of the specific analysis. While not containing a full sequence-of-events variable that would allow one to trace the entire crash sequence, GES does con- tain a number of variables that are of interest in this study, including a FHE and a MHE, both of which include striking a curb. The GES data do not contain any crash location infor- mation. Thus, they cannot be linked to any supplemental data, such as roadway inventories, operating speed inven- tories, or Average Annual Daily Traffic (AADT). National Automotive Sampling System— Crashworthiness Data System (NASS-CDS) The NASS Crashworthiness Data System contains detailed crash reconstruction data collected on site by expert investi- gators on approximately 5,000 crashes each year since 1979. 32 Data for the 1997–99 period were used in this study. The crashes must involve a vehicle that is towed from the scene. Thus, none of the less-severe property-damage-only crashes (the successes with respect to roadside objects) are included. The sample, taken from police reports in the same jurisdic- tions as the NASS-GES sample, is an unequal probability sample that is heavily weighted toward more severe crashes. The data are the highest quality crash data available, since they are based on detailed follow-up investigations by trained investigators. In order to develop national estimates from the data (or estimates related to the overall crash severity distri- bution), the cases have to be weighted based on the proba- bility of being selected. CDS includes a virtually unlimited “Sequence of Events,” with “Curb” as one of the objects that can be struck. Like the GES sample, there are no details of curbs or barriers, and only limited data on roadway geometrics; the data cannot be linked to supplemental roadway or traffic inventories. Unlike the GES data, the margin of error is rather large when one is exploring an issue with relatively few severe crashes per year (like curb-related crashes), since the sample for such crashes is quite small. Enhanced CDS data were developed at TTI for NCHRP Project 17-11, “Determination of Safe/Cost Effective Road- side Slopes and Associated Clear Distances.” NASS crash investigators collected additional data at selected CDS crash sites, and TTI reconstructed encroachment speed, angle, and tracking information where possible, including a confidence rating for the reconstructed data. FHWA’s HSIS HSIS is the only national data file containing both crash and roadway inventory elements. It includes linkable files of police-reported crashes, roadway geometry inventories, and traffic volumes in eight states (five states in the 1985–97 period; three additional states in the 1990–97 period). The files contain data for crashes of all severities on all state-system roadways, i.e., it excludes municipal or county roads not con- trolled by the state. Since the current project focused on higher-speed major roadways that the states control, this restriction was unimportant. While six of the eight states have some form of both “Object Struck” and “Sequence of Events” or “First/Most Harmful Event,” only Illinois and Michigan have a “Sequence of Events” variable in which curb impacts are separated from other objects and where “rollover” can be extracted as a separate event. Like the GES, both states also include information in the crash file related to crash/occupant injury severity and speed limit. Therefore, the data for these two states were chosen for use in this study. To capture the most recent years of data in the HSIS files, the 1996 and 1997 data for each state were used. A further advantage of HSIS is the linkable roadway inventory data. For both Illinois and Michigan, the inventory file includes not only AADT and speed limit for each section

of highway on the state system but also an indication of the presence of curb. Finally, Michigan provides an additional file not present in any other HSIS state: a Guardrail Inventory File that contains information on the location and description of each section (run) of guardrail along each side of the highway (e.g., type, purpose, and distance from roadway). Because there can be multiple rails at any point on the roadway (e.g., rails on each side and in the median), the file is very complex and difficult to work with. Furthermore, it has not been actively main- tained by the Michigan DOT since 1992. However, because this is the only known guardrail inventory file that can be linked with other roadway and traffic data to produce crash rates per passing vehicle, it was linked with the Michigan 1992 roadway inventory file and with Michigan 1993 and 1994 crashes in this study. Details of the complex merging effort and data decisions can be found in Appendix E. Description of Data Analyses Based on the goals of the project and the initial review of the available databases, the project panel defined a set of six 33 crash-data analyses to be conducted. Table 7 provides a brief description of each analysis along with the database used. As noted earlier, the analyses fell into two major groups: those conducted to further define and examine the extent of the curb-related safety problem, and those primarily conducted to provide input into the simulation and crash-testing efforts. Since an objective of the overall study effort was to relate curb design guidelines to some measure of roadway operat- ing speed (and, ultimately, to design speed), the panel was interested in targeting operating speed in the crash-data analyses where possible. Unfortunately, operating speed is captured neither in crash data nor in normal roadway inven- tory data. However, in a supplemental analysis, 1998 non- crash speed data were obtained from Michigan DOT and were used with New York State DOT data to define surrogate operating speeds for different combinations of functional class and speed limit. These surrogate operating speeds were then attached to crashes and used in the Michigan severity modeling effort and in the Michigan and Illinois rollover analyses. These operating speeds could not be used in other analyses due either to the nature of the issue (e.g., extreme crashes are a function of individual vehicle speeds rather than average roadway speeds) or to the source of the data (e.g., Task title Description Data used Extent of the U.S. Curb-Related Safety Problem The extent of the national safety problem related to curbs was documented. Questions addressed included, “how large is both the fatal and nonfatal crash problem, and has there been any trend over the past 5 years?” and “are there differences in the nature of the curb-related fatal and nonfatal crashes as compared to noncurb single-vehicle crashes?” 1994-99 FARS 1995-99 NASS-GES Examination of Curb-Related Rollover Risk and Nature Given a Crash Given the severity of rollovers in general and the nature of the curb, this was a detailed, multifile examination of the risk and nature of rollover given a curb-related crash. To help ensure that the curb was directly related to the rollover, all three databases chosen include a “sequence of events” that allowed selection of only rollovers preceded by a curb impact. 1997-99 NASS-CDS 1996-97 Michigan 1996-97 Illinois Crash, Injury, and Rollover Rates per Passing Vehicle for Guardrail Sections with and without Curbs To examine differences in the crash rates and rollover rates for guardrails with and without curbs, Michigan data on guardrail inventory, roadway inventory, traffic and crashes on urban freeway and other urban multilane roads were used in both contingency table analysis and negative binomial models. 1992 Michigan Guardrail Inventory and Roadway Inventory 1993-94 Michigan crash data Curb-Crash Severity Modeling To further examine curb-crash severity, Michigan data for SV crashes in which a curb was the first object struck and SV crashes in which no curb was struck were used in the development of ordinal regression models to examine the effect of crash-related variables (e.g., rollover, speed limit, weather, vehicle type, operating speed) on crash severity. 1996-97 Michigan Nature of Curb Impacts—Crash Reconstruction Data To provide guidance to crash testing and simulation efforts, an attempt was made to extract the specific nature of curb-related impacts (e.g., angle of impact, speed, tracking/nontracking) from both basic NASS- CDS data and from enhanced CDS data obtained from the Texas Transportation Institute. 1997-99 NASS-CDS TTI Enhanced CDS data Nature of Curb Impacts—Analysis Of “Extreme” Vs. “NonExtreme” Crashes Extreme and nonextreme (i.e., severe and nonsevere) curb crashes were compared to define crash conditions that differ between the two categories. Such identified conditions might provide both further basic information on curb safety and additional factors for consideration in simulation and crash-testing efforts. 1995-99 NASS-GES 1996-97 Michigan 1996-97 Illinois TABLE 7 Description of data analyses conducted and databases used

NASS-GES data do not contain functional class informa- tion). A more detailed description of the development of these assigned operating speeds is found in Appendix B. It should also be noted that since the goal of this project was to define curb guidelines for higher-speed roads rather than city streets, and as directed by the project panel, a speed limit of 40 mph (65 km/h) was used as the lower boundary for most of the analyses conducted; all exceptions are noted. COMPUTER SIMULATION METHODS As discussed in Chapter 2, computer simulation has been used to assess the safety effectiveness of curbs since the late 1960s. Many of these analyses were performed using HVOSM, a rigid body vehicle dynamics code. Although early computer programs were limited in their abilities (due in large part to computational constraints), the results of those analyses have provided a great deal of information regarding the effect of curb impact on vehicle kinematics. Vehicle dynamics codes have come a long way since the 1960s and are now able to provide very accurate results regarding vehicle kinematics. FEA is another computer simulation method that was use- ful in the study of curb and curb–barrier combinations. This method had not been used previously to study vehicle inter- action with curbs, but it has been used extensively in recent years to study vehicle impacts with roadside hardware. Since the early 1990s FEA has rapidly become a fundamental part of the analysis and design of roadside safety hardware sys- tems. In addition to being a reliable and relatively inexpen- sive means of analyzing and simulating impact events, it allows the analyst more control over the impact conditions and provides information about the mechanics of the impact event (stress, strain, energy, etc.) at specified time increments during impact. FEA is also capable of dealing with the highly nonlinear behavior associated with material properties, large deformations, and strain rate effects. The advantages and dis- advantages of using vehicle dynamics programs and FEA are discussed in the following sections. Vehicle Dynamics Codes The HVOSM is a vehicle dynamics program that has been used extensively in conjunction with full-scale crash testing to study vehicle dynamics during impact with curbs (14). Vehicle dynamics codes calculate the motions of the vehicle by modeling the vehicle as a series of rigid one-dimensional elements like springs, dampers, and masses. The tire and sus- pension models are the heart of a vehicle dynamics code since the only forces acting on the vehicle are presumed to arise from the tire interaction with the ground and inertia. The type of information that can be obtained from such analyses is related to the kinematics of the vehicle, such as vehicle tra- jectory, roll, pitch, and yaw. The trajectory of the vehicle has historically been used as a measure of the potential for over- 34 ride or underride of a barrier system. The HVOSM program has been modified and improved over the years and has been used for studying dynamic behavior of vehicles traversing various types of terrain. Development on HVOSM stopped, however, about 20 years ago as commercial vehicle dynam- ics codes supplanted it. HVOSM is now rarely used and vehi- cle suspension properties for modern passenger vehicles are not readily available for HVOSM. VDANL is a comprehensive vehicle dynamics simulation program that runs on a PC in a Windows environment (39). It was designed for the analysis of passenger cars, light trucks, articulated vehicles and multipurpose vehicles and has been upgraded over the years to expand and improve its capabili- ties. It now permits analysis of driver-induced maneuvering within limit conditions defined by tire saturation characteris- tics, as well as driver feedback control features. One of the significant advantages of using VDANL is that there is a large library of vehicle inertial and suspension properties available. Many of those properties have been validated by NHTSA using full-scale test track results. The one drawback of VDANL is that it is cannot simulate vehicle impact with an object and thus terrain must be smooth and continuous. This is because the program only simulates vehicle response due to interaction between the bottom of the tires and the ground. When a tire interacts with a curb that has a steep face, the contact will occur at a point higher up on the tire (i.e., not on the bottom of the tire), which cannot be accurately simu- lated with VDANL. Nonlinear, Dynamic Finite Element Codes For the simple event of vehicles traversing curbs, FEA provides little additional information about the kinematics of the vehicles than could be obtained through use of today’s vehicle dynamics codes. FEA was, however, invaluable in the analysis of impacts with curb–barrier combinations. Vehi- cle dynamics codes only provide information regarding vehi- cle kinematics and cannot provide information about the vehicle interaction with the barrier. The performance of traf- fic barriers installed in conjunction with curbs cannot be directly analyzed using vehicle dynamics codes, because they are not designed to account for deformations of the vehi- cle or barrier. Since vehicle dynamics codes only address suspension and inertial forces, they are not appropriate for use when a vehicle strikes a barrier. A vehicle striking a bar- rier experiences forces arising from the interaction of the vehicle body and the barrier itself. These forces are highly nonlinear and usually involve large deformations, plastic behavior, and, often, failure of materials. In FEA the entire substructure with its many parts and complicated shapes is divided into smaller units (finite ele- ments) that are interconnected at discrete points (nodes). The stresses, strains, and motions of the model are computed at the element level and are then combined to obtain the solu- tion of the whole body. The advantage of FEA is that the

body of the vehicle is not rigid, and thus it can deform in a realistic manner during impact, whether it be the simple elas- tic deformations involved in transferring the load through the framework of the vehicle when crossing curbs or the large, plastic deformations involved in vehicle impacts with road- side safety barriers. Vehicle dynamics codes have been used in previous stud- ies to determine the potential for vaulting over or underriding barriers. In those studies, however, such potential was only speculated based on the vehicle’s trajectory after crossing a curb; an actual impact event is much more complicated. FEA can provide detailed information about the impact event, including vehicle kinematics prior to and during interaction with the barrier, as well as damage sustained by both the vehicle and the barrier. FEA can also provide vehicle accel- eration data that can be used for measuring injury risk factors of vehicle occupants. For many years, full-scale crash testing was the primary method of determining the effectiveness of roadside safety hardware. More recently, there has been a great deal of advancement in computation power and in code develop- ment (40). As a result the use of FEA for simulating collision events has become a reliable and widespread tool for inves- tigating crashworthiness of roadside safety structures. In 1998, the FHWA began the Centers of Excellence Pro- gram, in which it funds leading research organizations, including Worcester Polytechnic Institute (WPI), to investi- gate the impact performance of various roadside safety hard- ware. LS-DYNA was chosen by the FHWA to serve as the primary analysis tool to be used by the centers. LS-DYNA is a nonlinear, dynamic, explicit finite element code that is very efficient for the analysis of vehicular impact and is used extensively by the automotive industry to analyze vehicle crashworthiness (41). It evolved from DYNA3D, public domain software developed in the mid- to late 1970s by John Hallquist at Lawrence Livermore National Laboratory. LS- DYNA’s efficiency in simulating contact between various parts in a finite element model, along with its ability to effec- tively use underintegrated elements, has put LS-DYNA at the forefront of the nonlinear dynamic finite element soft- ware industry. One advantage of FEA is that it is easy to vary parameters and assess exactly the structural and dynamic context of the 35 collision. Parametric analyses are particularly straightforward, using simulation so that the variation of speeds and angles can be examined to find the critical impact conditions at which poor performance might occur. Simulation provides a method to explore a wide variety of curb–barrier combina- tions that would provide the broadest type of information for development of guidelines for the use of curb or curb– barrier combinations. The primary drawback of finite element simulations is that they must be validated to make sure that the predictions are realistic. There are several public domain vehicle models available from the FHWA/NHTSA National Crash Analysis Center at George Washington University that have been validated for various impact conditions. A list of currently available vehi- cle models appears in Table 8. Of the vehicle models listed in the table, the 1994 Chevro- let C-1500 reduced model has been used most widely by WPI researchers in particular and the Centers of Excellence com- munity in general. While any of the models listed in Table 8 could have been used in this project, there is often consider- able work needed to make a model useable in a particular impact scenario. The 1994 reduced model of the Chevrolet C-1500 was the easiest model to use since it had been widely used and debugged. The 1994 Chevrolet C-1500 (detailed model) and the 1993 Ford Taurus were also reasonably debugged but most of the other models had not been widely used outside of the NCAC and might have required signifi- cant debugging to be useful in this research. The basic procedure used by the researchers at WPI in pre- vious projects using FEA to examine roadside hardware has three steps: (1) build the finite element models, (2) validate them using crash tests found in the literature, and then (3) use the validated models to develop alternative designs. This pro- cedure was followed in this project to ensure that the guide- lines were based on models that had been validated against observable physical phenomena (e.g., crash tests). Validation of Computer Models Computer simulations were validated by comparing the simulated results to those obtained from full-scale crash tests. The accelerations at the center of gravity of the vehicle in the Vehicle model type 1998 Oldsmobile Cutlas Ciera 1996 Ford F-Series Truck 1994 Chevrolet C-1500 (detailed model) 1997 Geo Metro 1994 Chevrolet C-1500 (reduced model) 1993 Ford Taurus 1996 Plymouth Neon Honda Accord Chevrolet Lumina Dodge Intrepid Ford Crown Victoria Ford Explorer TABLE 8 Public domain vehicle models available from the National Crash Analysis Center

simulation and the full-scale test were compared using four quantitative techniques: 1. the Numerical Analysis of Roadside Design (NARD) validation parameters, 2. the analysis of variance (ANOVA) method, 3. the Geers parameters, and 4. the Test Risk Assessment Program (TRAP). The NARD validation procedures are based on concepts of signal analysis and are used for comparing the acceleration- time histories of finite element simulations and full-scale tests (42). The ANOVA method is a statistical test of the residual error between two signals (43). Geers’ method compares the magnitude, phase, and correlation of two signals to arrive at a quantitative measure of the similarity of two acceleration- time histories (44). TRAP is a software program that was developed to evaluate actual full-scale crash tests and gener- ate important evaluation parameters like the occupant impact velocities (OIVs), ride down accelerations, 50 msec average acceleration, and so forth. The program calculates standard- ized occupant risk factors from vehicle crash data in accor- dance with the NCHRP guidelines and the European Commit- tee for Standardization (CEN) (45). Using the same evaluation software for finite element simulations and full-scale tests further simplified the comparisons between actual physical tests and mathematical simulations. Applicability of FEA to Roadside Barrier Impact Studies Researchers at WPI had considerable experience using the LS-DYNA program for simulating vehicle impacts into roadside hardware (46). As part of previous FHWA projects, Plaxico and Ray had developed finite element models of vari- ous roadside structures that were used to assess the impact per- formance of the systems. All the models were validated with the results of full-scale crash tests (31). These models included the breakaway cable terminal; the MELT terminal; a weak- post guardrail system; and two strong-post guardrail systems, the G4(1W) and G4(2W) (47–50). The G4(1W) and G4(2W) are both blocked-out strong-post W-beam guardrails; the G4(1W) uses 200 x 200mm wood posts; and the G4(2W) uses 150 x 200 mm wood posts. The G4(1W) is used in Iowa, and the G4(2W) is used in a number of other states. A finite element model of the G4(2W) guardrail had been developed by researchers at WPI as part of a study sponsored by the Iowa Department of Transportation and the FHWA (46). Simulations of Report 350 Test 3-11 impact conditions were performed with the model, and the results were com- pared to a full-scale crash test performed by TTI that estab- lished that the guardrail system successfully met the stan- dards set in NCHRP Report 350 (31). Figures 21 and 22 compare the FEA to the results of the full-scale crash test. This model was validated using the methods described pre- 36 viously. There was good agreement between the test and the simulation with respect to velocity histories, event timing, exit conditions, guardrail damage, and guardrail deflections, as well as the TRAP, NARD, Geers, and ANOVA evaluation parameters. A summary of major impact events, the time at which they occurred, and the corresponding velocity of the vehicle is presented in Table 9. Both the qualitative and quantitative comparisons of the finite element simulation to the physical crash test indicate that the simulation results rea- sonably replicate the guardrail performance in the test. As an example of the use of FEA in this project, the vali- dated model of the G4(2W) was used to simulate a Test Level 3 impact event involving the G4(2W) with a 150-mm-high AASHTO Type B mountable curb located just behind the face of the W-beam. The results are shown in Figure 23. The impact conditions were the same as those in TTI Test 471470-26. A rear view of both of the simulations (i.e., with and without a curb) is compared in Figure 24. From the results of the simulations it appears that the 150-mm-high AASHTO Type B curb placed behind the face of the G4(2W) guardrail system will likely cause serious instability when the vehicle exits the system. It is commonly observed in full- scale tests involving the 2000-kg pickup truck impacting var- ious roadside barriers that when the rear tire contacts the bar- rier, the rotation of the tire tends to pitch the rear of the vehicle upwards, as shown in Figure 21. This phenomenon is further amplified when a curb is placed in combination with the guardrail. When the rear wheel hits the curb, an initial vertical displacement of the wheel prior to tire interaction with the barrier results, as demonstrated in Figures 23 and 24. The high pitch and exit angle of the vehicle during impact with the curb–guardrail combination make the post-impact behavior of the pickup very unpredictable. Rollover would be very likely given the exit conditions shown in Figure 23. Typically, during impact with strong-post guardrail sys- tems without a curb present, the front wheels of the pickup truck remain in contact with the ground over much of the event, which in effect reduces the lateral deflection of the system during impact and also decreases the redirection angle of the vehicle as it exits the system. In this finite ele- ment simulation of the curb–guardrail combination the vehi- cle was completely airborne during the time that it was in contact with the barrier, resulting in increased lateral deflec- tion of the barrier and a much higher angle of redirection of the vehicle. The total deflection of the system in the simula- tions with and without a curb was 0.79 m and 0.71 m, respec- tively (i.e., the deflection in curb–barrier combination was 11.2% greater). The redirection angles of the vehicle in the simulations with and without a curb were 14 and 21 degrees, respectively. The redirection angle of the vehicle in the curb– guardrail simulation exceeded the allowable exit angle spec- ified in NCHRP Report 350. According to criteria M of Report 350, the exit angle from the test article should be less than 60% of the test impact angle, measured at time of vehi- cle loss of contact with test device. The exit angle in the curb–guardrail simulation was 84% of the impact angle.

37 Figure 21. Sequential photographs for TTI Test 471470-26 (left) and G4(2W) finite element simulation (right).

PARAMETRIC ANALYSES USING COMPUTER SIMULATIONS As demonstrated in the simulations, the potential for either barrier failure or vehicle vaulting can be assessed in much the same way that physical crash tests are evaluated. The advan- tage of finite element simulations is that once a model is developed and validated, the impact conditions, as well as the basic geometry of the installation, can be varied easily. Per- forming ten finite element simulations with the curb located at different distances from the face of the post, for example, would be straightforward and inexpensive and would allow the analyst to determine the effect of the curb offset on the performance of the barrier. Likewise, curbs with heights vary- ing from 0 to 300 mm could be evaluated easily using finite 38 element simulations. Another variable that could be inves- tigated is vehicle speed. It is of interest to highway engi- neers to know the maximum impact speed that a system can withstand. Such information could be used for determining which system would be the most effective along a given stretch of roadway where site and operating conditions are known. Due to the fact that the project had limited funds, the proj- ect team and panel had to balance the number of simula- tions with the number of possible scenarios that could be investigated. Analysis of Curb–Barrier Combinations Analyses involving curb–barrier combinations were per- formed using the LS-DYNA finite element software. A matrix Time = 0.061 secondsTime = 0.000 seconds Time = 0.119 seconds Time = 0.180 seconds Time = 0.241 seconds Time = 0.361 seconds Time = 0.480 seconds Time = 0.600 seconds 15 14161718 Figure 22. Sequential photographs for TTI Test 471470-26 (top) and G4(2W) finite element simulation (bottom), overhead view.

G4(2W) Full-scale test Finite element simulation Summary of impact events Time (sec) Speed (km/h) Time (sec) Speed (km/h) Initial Contact 0.000 100.8 0.000 100.8 Vehicle starts to yaw 0.056 100.8 0.044 100.6 Wheel impacts post 15 0.104 90.2 0.101 91.3 Wheel impacts post 16 0.193 74.8 0.190 75.7 Rear of vehicle contacts guardrail 0.203 73.2 0.207 73.0 Wheel Detaches 0.215 69.4 0.215 71.3 Vehicle parallel with guardrail 0.283 68.0 0.264 69.0 Vehicle exits guardrail θ = 13.5E 64.0 θ = 14.3E 63.0 TABLE 9 Summary of major impact events of test 471470-26 and G4(2W) finite element simulation (46 ) Front View Overhead View Figure 23. Finite element simulation of a 2000P vehicle striking a G4(2W) with a 150-mm-high AASHTO Type B mountable curb.

of simulations was developed to provide information regard- ing the impact performance of the G4(1S) guardrail system in combination with various types of curbs at impact conditions specified by NCHRP Report 350 Test 2-11 and Test 3-11. Both these tests involve the 2000-kg pickup truck impacting at 25 degrees. The impact speed for Test 2-11 is 70 km/h, which is in the intermediate speed range (i.e., 60 to 80 km/h), and the impact speed for Test 3-11 is 100 km/h, which rep- resents the higher speed range (i.e., > 80 km/h). The perfor- mance of certain curb–barrier systems was also investigated at 85 km/h, which represented the upper limit of intermedi- ate speed roadways (i.e., 60-80 km/h). There are many barrier systems that could have been investigated in the study, such as the G4(2W), G9 (thrie- 40 beam), G2 (weak-post W-beam), or G1 (weak-post cable), but it was decided to investigate combinations of curbs with the more widely used systems. The G4(2W) and the modi- fied G4(1S) (i.e., steel posts with wood blockouts) are widely used systems and were good candidates for the research. Since both systems have successfully passed NCHRP Report 350 TL-3 impact conditions, poor performance of these sys- tems combined with a curb can be directly attributed to the presence of the curb and not necessarily to structural inade- quacy of the barrier systems. Since there were a limited num- ber of analyses that could feasibly be conducted, only the modified G4(1S) guardrail was used in the study so that the maximum number of curb types and impact conditions could be investigated. The G4(1S) is the most widely used strong- G4(2W) with Curb G4(2W) without Curb Figure 24. Sequential photographs of finite element simulations comparing the impact performance of the G4(2W) with and without the AASHTO Type B curb.

post guardrail in the United States, thus information regard- ing its performance with curbs should be the most beneficial to the states. NCHRP Report 350 Test 2-11 and Test 3-11 impact condi- tions were chosen for the matrix of simulations because they involve the 2000-kg pickup, which is much more unstable than the 820-kg small car and also produces a more severe impact due to the larger mass of the pickup. The simulations were used to determine the most effective curb–barrier combinations for those impact conditions. Analysis of Vehicle Impacts with Curbs Analyses involving the simple impact of a vehicle and curb were also investigated using LS-DYNA. There are a number of variables that would have been interesting to investigate in this study, such as vehicle type (e.g., small car, pickup, SUV), curb type, impact speed, and angle of impact. Due to limita- tions in time and computational constraints, only a limited number of impact conditions were investigated. A matrix of simulations was developed to provide information regarding the vehicle’s response when crossing a number of different curb types at various impact conditions. The information col- lected in this phase of the study served two purposes: (1) to quantify the effects of vehicle impact with curbs on the sta- 41 bility of the vehicle and (2) to provide information regarding the trajectory and path of the vehicle after impact with curbs. Most of the curb impact studies that were identified in the literature involved vehicles encroaching the curb in a track- ing manner. Another aspect of collisions with curbs involves an out-of-control vehicle impacting the curb in a nontracking position. In these situations, vehicle tripping may be highly probable during impact. Nontracking impacts with curbs may result in vehicle instability and rollover, especially impacts involving vehicles with high centers of gravity. The side friction between the tires and ground for an out- of-control vehicle will cause the vehicle to roll, such that the vehicle has an initial roll-rate at the onset of impact with the curb. This factor is much more significant for vehicles with a high center of gravity, such as pick-up trucks and SUVs which make up a large percentage of the vehicle population currently on the road. As documented in NHTSA’s Rollover Status Report in Traffic Safety Facts 1996 (51), rollover crashes, particularly single-vehicle (SV) accidents in light pickup trucks and SUVs, continue to take the lives of thousands of Americans each year. In 1996, almost 9,500 passenger vehicles (e.g., passenger cars, pickup trucks, vans, and SUVs were involved in fatal rollover crashes. Rollovers accounted for 36% of all fatal crashes involving SUVs and 24.5% of all fatal crashes involving pickup trucks, as illustrated in Figure 25. It is also Figure 25. Rollover occurrence as a percent of all crashes, by vehicle type and crash severity (51).

notable that 5.3% of all accidents involving SUVs resulted in rollover. The large percentage of SUVs and pickup trucks on today’s highways along with their high rollover rate make nontracking impact with curbs a much more important factor now than in former years. There has been a great deal of advancement in computation power and in code develop- ment over the past few years that has enabled computer sim- ulation programs to become a very efficient means of analy- sis. Although nontracking simulations were not included in this project, in theory both tracking and nontracking impacts with curbs could be investigated using a vehicle dynamics code, such as VDANL. FULL-SCALE CRASH TESTING Introduction Full-scale crash testing is the method used by the FHWA to certify that a barrier system is crashworthy for use on federally funded highways. Although advancements in computer sim- ulation programs have made it possible to accurately repro- duce and predict complex impact events, full-scale testing is still essential in evaluating the safety performance of roadside appurtenances, including curbs and curb–barrier systems. To evaluate the performance of roadside safety barriers, impact conditions must meet the standard testing procedures accepted by the FHWA. The current procedures are published in NCHRP Report 350. Prior to Report 350, the 2040-kg pas- senger sedan served as the crash test vehicle representing the large end of the passenger vehicle fleet. Because the large passenger sedan had virtually disappeared from the vehicle population by the late 1980s and new vehicle types, such as minivans, SUVs, and pickup trucks, had emerged in its place, Report 350 replaced the large car with a 2000-kg pickup truck. The pickup truck introduced new challenges in crash testing due to its high center of gravity, which makes it much more unstable during impacts than the large car. The 2000-kg pickup truck was chosen as a replacement for the 2040-kg passenger sedan for several reasons. First, both vehicles had similar mass and were therefore thought to rep- resent a similar barrier loading. Second, the pickup truck was chosen as a surrogate for a much broader class of vehicles. The Intermodal Surface Transportation Efficiency Act of 1991 (ISTEA) required the FHWA to address the issue of the crashworthiness of the emerging SUV fleet; the FHWA responded by adopting the 2000-kg pickup truck in Report 350 as a surrogate for the entire class of SUVs (e.g., pickup trucks, SUVs, minivans, and vans), now known as ISTEA vehicles. While some of the small SUV vehicles have worse stability characteristics, the pickup truck is one of the least stable vehicles in the vehicle fleet. It is characterized by a high center of gravity positioned far forward in the vehicle. There is little front overhang and the suspensions are rela- tively stiff. Testing with the pickup truck has presented some difficult challenges because of its inertial and stability char- 42 acteristics. In the context of developing guidelines for curbs and curb–barrier combinations, it is important to remember that the pickup truck is not only an important test vehicle in its own right but also a surrogate for the broader class of ISTEA vehicles. The performance of a curb–guardrail combination can be evaluated using test conditions specified in NCHRP Report 350 for evaluating the crashworthiness of the length of need section of a longitudinal barrier. There are currently two tests required to evaluate guardrail systems for TL-3: 1. Test 3-11, in which a 2000P pickup truck (e.g., Chevro- let 2500) impacts the guardrail at a speed of 100 km/h and impact angle of 25 degrees, and 2. Test 3-10, in which an 820C (e.g., Honda Civic or Ford Festiva) impacts the guardrail at a speed of 100 km/h and impact angle of 20 degrees. A guardrail system that meets all the strength and safety requirements specified in NCHRP Report 350 is considered acceptable for use on all federal-aid roadways within the United States. The literature review identified a limited number of full- scale tests involving vehicle impacts with curbs and curb– guardrail combinations. While full-scale crash testing was used in almost every study that involved vehicle-curb impact, all the tests that involved simple vehicle-to-curb impacts were performed using a large 2040-kg passenger sedan. The results of those earlier tests may have little significance regarding the effects of curb impact with the current fleet of vehicles, which ranges from very lightweight compact cars to large, unstable pickup trucks and SUVs. In this project, a full-scale testing program was used to ver- ify and confirm hypotheses developed from the computer sim- ulations and to validate and strengthen the conclusions of the parametric studies. The few full-scale tests of curb–barrier combinations that were identified in the literature aided in the validation of the models so that the number of additional tests could be minimized. Low-Speed Curb Traversal Tests Full-scale live-drive tests were performed on three dif- ferent types of curbs (AASHTO B curb, G curb, and verti- cal 6-in. curb) at varying speeds and angles (10, 15, 25, and 90 degrees). The test area was a gravel parking lot. The curbs were made using reinforced concrete cast in 1.2-m-long sec- tions. Each set of curbs was attached to the ground with steel rods driven through holes in the curbs into the gravel. The area behind the curb was backfilled with gravel up to the top of the curb. The test setup is shown in Figures 26 and 27. The vehicle path was marked on the ground using plastic strips. The driver aligned the vehicle with the strips to attain the desired approach angle, accelerated the vehicle to the desired speed, and then released the steering wheel just prior to striking the curb. After the rear wheels crossed the curb,

the driver reasserted control of the vehicle by steering and applying the brakes. Each test was performed multiple times to assess the repeatability of the event. The relative displacements of all four wheels and the accel- erations in the longitudinal, lateral, and vertical directions at two points on the vehicle were measured during each test. 43 Moderate-Speed Live-Driver Tracking Tests of AASHTO Mountable Curbs Full-scale curb traversal tests were next performed at mod- erate speeds (i.e., approximately 56 km/h) with a live driver. The purpose of the tests was to evaluate the trajectory and Figure 26. Full-scale curb test setup. Figure 27. Full-scale curb test setup—overhead view.

kinematics of a typical 2000P vehicle traversing different types of AASHTO curbs at higher speeds. Due to the physi- cal limitations of the testing site, only the low end of the speed range of interest (60 to 100 km/h) could be tested safely. During these tests, data were collected about the dis- placements and accelerations experienced by the vehicle. The tests were performed using the 1995 Chevrolet C2500 Cheyenne pickup truck shown in Figure 26. The truck was modified by removing the bed and installing a roll bar, anti- rollover outriggers and ballast weights. The final mass of the vehicle, ready to be tested and refueled, was 2,165.90 kg; the final mass with fuel and driver was 2,248.00 kg. The truck was driven toward a 12-m-long curb installation at angles of 15 and 25 degrees. Since the test vehicle was controlled by a driver, it was difficult to obtain precise, repeat- able impact conditions. The driver was instructed to follow a painted line on the testing area and to hit the curbs at 35 mph (15.65 m/s). Due to the runway length available and the vari- ability due to human and vehicle performance, the actual impact speed varied. After each test, the driver reported the impact speed. Brakes were applied by the driver only after the vehicle had crossed the curb. Moderate-Speed Live-Driver Nontracking Tests of AASHTO Mountable Curbs Nontracking full-scale curb traversal tests were also per- formed at moderate speeds (i.e., approximately 56 km/h) with a live driver. The purpose of these tests was to evaluate the vehicle trajectory and kinematics of a typical 2000P vehicle traversing different types of AASHTO curbs in nontracking mode in order to investigate the extent to which the curbs act as a tripping mechanism for vehicle rollover. These tests were performed using the same 1995 Chevro- let C2500 Cheyenne pickup truck with the same modifica- tions as for the tracking tests. The driver executed two dif- ferent maneuvers resulting in a nontracking impact with the curb. These maneuvers were intended to reproduce two typ- ical scenarios of vehicles running off the roadway, over- steering and understeering. In scenario 1, oversteering, the vehicle was accelerated to a constant velocity of 35 mph (56 km/h) in a straight-line tra- jectory at a 55 ± 10( angle with respect to the curb line. At a marked point 6 m before the curb line, the driver turned the steering wheel approximately 45 degrees and immediately activated the emergency brake (i.e., rear brakes only) to break loose the rear end of the vehicle. In scenario 2, understeering, the vehicle was accelerated to a constant velocity of 35 mph (56 km/h) in a straight-line tra- jectory at a 55 ± 10( angle with respect to the curb line. At a marked point 6 m before the curb line, the driver turned the steering wheel to approximately 60 degrees without applying the brakes. 44 For both scenarios, the truck impacted a 12-m-long instal- lation of AASHTO curbs. The tests were conducted using curb types B, C, D, and NY. Full-Scale Crash Tests of Curb–Guardrail Combinations Several full-scale tests were conducted of 2500P trucks impacting curb–guardrail combinations. The test reports are included in Appendix I of this report. The impact conditions were similar to NCHRP Report 350 Test 3-11. The follow- ing articles were tested: • AASHTO Type B curb directly beneath modified G4(1S) guardrail, • AASHTO Type B curb positioned 2.5 m in front of mod- ified G4(1S) guardrail, and • New York Type T100 curb positioned 4.5 m in front of modified G4(1S) guardrail. The test vehicles were a 1998 GMC 3/4-ton pickup (test iner- tial mass of 1,993 kg), 1994 Chevrolet 3/4-ton pickup (test iner- tial mass of 2,002 kg), and a 1989 GMC 3/4-ton pickup (test inertial mass of 2,014 kg). The guardrails tested were 53.34-m installations of AASHTO SGR04a guardrail with a SEW02a End Terminal and Re-Block recycled plastic blockouts made of 50% high-density polyethylene and 50% polypropylene. They were installed in dry NCHRP Report 350 Strong Soil. Figure 28 shows the test vehicle and configuration for the curb directly beneath the guardrail. In each test, the vehicle impacted the curb at approxi- mately 85 km/h and 25 degrees. The critical impact point was near the midpoint of the guardrail installation, 0.6 m upstream of Post 14 and 2.5 m upstream of a connection splice respectively. SUMMARY Real-world crash data were used to better characterize safety problems associated with curb and curb–barrier com- binations on higher-speed roadways and to provide leads to the crash testing and simulation efforts conducted in this proj- ect. The analyses conducted with crash data included the fol- lowing: assessment of the extent of the U.S. curb-related safety problem; examination of curb-related rollover risk and nature given a crash; comparison of crash, injury, and roll- over rates per passing vehicle for guardrail sections with and without curbs; curb-crash severity modeling; and examina- tion of the nature of curb impacts, using crash reconstruction data and comparing extreme and nonextreme crashes. FEA was also used to study the effects of vehicle inter- action with curbs and curb–guardrail combinations. The advan- tage of computer simulation is that once a model is developed,

the impact conditions and the basic geometry of the installa- tion can be varied easily. The finite element program LS- DYNA was used in a parametric study to investigate the response of vehicles crossing various types of curbs. LS- DYNA was also used to investigate the effects of installing curbs in conjunction with guardrail, regarding the ability of the barrier to safely contain and redirect an impacting vehicle. 45 Full-scale crash tests were used to validate the computer models. Live-driver curb traversal tests were performed at low and moderate speeds in tracking and nontracking modes. Several full-scale tests of curb–guardrail combinations were also performed at higher speeds. The results of these analyses are discussed in Chapter 5 of this report. Figure 28. Test vehicle and setup for Type B curb beneath guardrail.