A similarity scoring technique to analyse comparisons of real-world crashes to crash tests: initial results from a 12-point system

Transcription

1 Int. J. Vehicle Safety, Vol. 6, No. 3, A similarity scoring technique to analyse comparisons of real-world crashes to crash tests: initial results from a 12-point system Kathryn L. Loftis Virginia Tech-Wake Forest University Center for Injury Biomechanics, School of Biomedical Engineering and Sciences, Medical Center Blvd., Winston-Salem, North Carolina, 27157, USA kaloftis@wakehealth.edu R. Shayn Martin and J. Wayne Meredith Wake Forest University School of Medicine, Watlington Hall, General Surgery Dept., Medical Center Blvd., Winston-Salem, North Carolina, 27157, USA romartin@wakehealth.edu merediw@wakehealth.edu Joel D. Stitzel* Virginia Tech-Wake Forest University Center for Injury Biomechanics, School of Biomedical Engineering and Sciences, Medical Center Blvd., Winston-Salem, North Carolina, 27157, USA jstitzel@wakehealth.edu *Corresponding author Abstract: The most similar crash test was identified for each of 100 Crash Injury Research and Engineering Network (CIREN) cases. To quantify the best comparison pairs, a Similarity Scoring Methodology (SSM) with 12 parameters was developed. The results showed 18 comparisons with low (0 6 points), 72 with medium (7 9 points) and 10 with high (10 12 points) similarity scores. Thirty-nine CIREN cases received a similarity point for deltav (within range: ±16.1 kph [10 mph]). Thirty-seven CIREN cases had a lower deltav than the crash test. For occupant parameters, seating position and airbag deployment received similarity points most frequently (86% each). Occupant height and weight received points least frequently (41% and 20%, respectively), typically because CIREN occupants were shorter and heavier compared with Anthropometric Test Device (ATD) sizes. This work establishes a standard SSM to be used with future studies and provides information about key differences between crash tests and real-world crashes. Keywords: motor vehicle crash; similarity scoring; vehicle safety; CIREN; Crash Injury Research and Engineering Network; NHTSA; National Highway Traffic Safety Administration; IIHS; Insurance Institute for Highway Safety; crash comparisons. Copyright 2013 Inderscience Enterprises Ltd.

2 192 K.L. Loftis et al. Reference to this paper should be made as follows: Loftis, K.L., Martin, R.S., Meredith, J.W. and Stitzel, J.D. (2013) A similarity scoring technique to analyse comparisons of real-world crashes to crash tests: initial results from a 12-point system, Int. J. Vehicle Safety, Vol. 6, No. 3, pp Introduction Rigorous safety testing is performed in many ries around the world to ensure that occupants remain safe when their vehicles are involved in crashes. These crash tests lead to new safety designs, including seatbelts, crush zones and airbags. While vehicle safety has been greatly improved by these crash tests, serious injuries and fatalities are still seen in Motor Vehicle Crashes (MVCs). In 2008, more than 37,000 deaths and 2.3 million injuries were reported in United States (US) motor vehicle crashes (NHTSA, 2010b). Studying current injuries provides better information about vehicle safety and crash tests, and by continually improving crash tests and introducing new vehicle safety features, more lives can be saved. The Crash Injury Research and Engineering Network (CIREN), a national program funded by the National Highway Traffic Safety Administration (NHTSA), collects information about occupants injured in motor vehicle crashes. CIREN teams conduct case reviews to determine the mechanism for each injury using scene evidence, interior vehicle evidence, crash severity measures and occupant information. Through CIREN, multidisciplinary researchers at six Level 1 Trauma Centers across the US conduct indepth studies of crashes resulting in serious injury with the goal of improving vehicle safety, trauma triage processes and patient outcomes. 1.1 Crash tests The NHTSA and Insurance Institute for Highway Safety (IIHS) in the US conduct multiple tests to simulate real-world crashes under various conditions. Common Federal Motor Vehicle Safety Standard (FMVSS) tests from NHTSA include FMVSS 208, a full frontal barrier test with unbelted anthropomorphic test devices (ATDs, i.e. crash test dummies); and FMVSS 214, a side impact moving deformable barrier crabbed test. NHTSA New Car Assessment Program (NCAP) tests are run at higher speeds than the standard FMVSS tests and all ATDs are belted, therefore these tests provide a different set of parameters for comparisons (NHTSA, 1990, 2007, 2008a, 2008b). The IIHS is a consumer agency that performs vehicle crash tests different from NHTSA s. The IIHS side impact crash tests use a larger barrier than the NHTSA crash tests, to better represent the front of a sport utility vehicle (IIHS) (Summers et al., 1999). Vehicle incompatibility is shown to generate more injuries for the occupants in the smaller vehicle, especially in side impacts where the front bumper of the larger vehicle impacts above the level of the door sill of the smaller vehicle (Gabler and Hollowell, 1998; Summers et al., 2001; Loftis et al., 2009). The IIHS also performs frontal crash tests using a 40% offset frontal stationary barrier instead of a full frontal configuration, with an ATD belted in the driver s seating position. The test simulates an offset frontal impact where a vehicle departs the travel lane to the left and impacts an oncoming vehicle (IIHS).

3 A similarity scoring technique 193 ATD measurements recorded during crash tests are used with injury criteria by NHTSA and IIHS to assess a vehicle s crashworthiness (IIHS; IIHS, 2008; NHTSA, 2008c). Fiftieth percentile male Hybrid III ATDs have been included in testing for FMVSS 208, frontal NCAP, and IIHS offset frontal crash tests (IIHS; NHTSA, 2004; US GOA, 2005). The ATD most commonly available in previous NHTSA side impact crash tests (FMVSS 214 and side NCAP) is the Side Impact Dummy (SID) 50th percentile male, which was used with vehicle model years 1997 through 2010 (NHTSA, 1990; Kahane, 1999; NHTSA, 2007). IIHS side impact crash tests use the SID-IIs 5th percentile female ATD (Build Level C) and have been conducted on vehicle model years starting with 2002 through 2011 (IIHS; IIHS). 1.2 Comparison studies Previous studies have researched comparisons between crash tests and real-world crashes. These previous comparison studies were often based on data ranges for defining similarity between parameters, including deltav, maximum crush and Principal Direction of Force (PDOF) (Ryan and Mackay, 1969, UMPIRE). Council et al. created a comparison method for predicting the crashworthiness of new vehicles based on older models of the same make using a vehicle clones list and vehicle size categories (Council et al., 1997). The current study uses a similar approach to identify vehicles within the crash test databases that are applicable to real-world crash vehicles. Tencer et al. (2005) analysed data from the US NCAP, National Automotive Sampling System- Crashworthiness Data System (NASS-CDS) and CIREN data to examine pelvic and thoracic forces in side impacts. This Tencer et al. s study demonstrated methods for comparing real-world crash data with crash test data. Tencer et al. also showed a correlation between reduced Thoracic Trauma Index (TTI) injury risk and reduced injury severity seen in real-world crashes, but NCAP test results underestimated thoracic injury (Tencer et al., 2005). Previously, a Mahalanobis distance technique similarity scoring system was developed by Yu et al. to compare two real-world crash databases (Yu et al., 2008). In 2010, Loftis et al. described steps necessary for finding crash tests similar to real-world cases using variables available in both and provided an example comparison case. Initial comparisons were achieved by qualitatively analysing crash tests and realworld cases to find the most similar comparisons (Loftis et al., 2010). The basis for the current paper originates from the 13 important crash and occupant parameters that were identified in the previous paper. The objectives of this current study were to build on the methodology described by Loftis et al. (2010) and develop a quantitative Similarity Scoring Methodology (SSM) for identifying the best comparison pairs. First, the best crash test for comparison to each real-world crash is identified through a sequential selection process. Then the comparison pairs are scored using 12 similarity parameters. After scoring each comparison, the similarity parameters are analysed to determine common factors between real-world cases and crash tests. This research provides the first step in developing a standard SSM for researchers to build on for future comparisons between crash tests and real-world cases and can be applied to studying real-world cases in comparison to other ries crash tests by altering the crash test conditions. By identifying the best comparisons of real-world cases to crash tests, it is possible to further investigate why injury may have occurred in the real-world case.

4 194 K.L. Loftis et al. 2 Methods NHTSA and IIHS crash test databases were used for comparison with real-world crashes in the Wake Forest University (WFU) CIREN centre database. WFU CIREN cases were used because full patient radiology was available for each injury. This allows future researchers to investigate occupant injuries for the comparisons. s 001 through 120 from the WFU CIREN centre were selected for analysis; 20 cases with non-horizontal main impacts were excluded, leaving 100 for comparison with crash tests. The first portion of the methods (Part 1) focuses on identifying the best crash test comparison for each real-world case using sequential parameters. The most important parameter for the comparisons was the crash type either frontal or side. After separating cases by the crash type, 12 specific parameters were investigated. These included the crash configuration (i.e. frontal offset or near-side impact and speed at impact), type of vehicle and occupant characteristics (i.e. height, weight, seating position, belt use and airbag deployment). The selection of the similar crash test for each real-world case is discussed in more detail in the first section. The second step identifies the best pairs of comparisons for further analysis (Part 2). The SSM was developed to quantify comparisons between real-world cases and crash tests using the 12 parameters identified (Loftis et al., 2010). The first parameter (crash type) from the previous study was used to identify the cases that would be scored. The 12-parameter SSM was used to quantify the comparison of the real-world crash to the crash test, for each case. The parameters were equally weighted, at one point each, resulting in a maximum score of 12 points. Each parameter was given equal weight in this initial study to establish a baseline to assess future enhancements to the scoring methodology. 2.1 Part 1: selection of similar crash test For each of the 100 CIREN cases, crash, vehicle and occupant characteristics were examined in turn to determine which crash test database to search for a comparison test. These methods were first provided in a previous study, where Loftis et al. described important comparison parameters available in both real-world and crash test databases (Loftis et al., 2010). Table 1 shows the most frequently used crash tests for comparisons in this study. Real-world cases were first identified by designating them as either frontal or side crashes (based on the highest severity impact). Because vehicle dynamics play a large role in occupant kinematics and injury patterns, crash parameters were investigated first in the sequential methodology for crash test selection. For frontal crashes the next parameters examined were crash type (offset frontal IIHS, full frontal NHTSA, pole), occupant seating position (driver or other passenger [no IIHS comparison for occupant other than driver]), deltav (high like NCAP, IIHS or low like FMVSS 208), occupant size (more like 5th percentile female, 50th percentile male or paediatric case) and belted status of the CIREN case occupant. These dictated whether to search in the IIHS database or the NHTSA database for a comparison test crash. For side impacts, the occupant size (more like 5th percentile female or more like 50th percentile male), striking vehicle size (car-nhtsa; or SUV-IIHS) and deltav of the CIREN case vehicle dictated which database to search. While the most common crash tests used for this analysis were barrier tests, pole tests were researched where applicable. Pole impacts were scarce both among CIREN cases and in the crash test databases.

5 A similarity scoring technique 195 Table 1 Common crash tests used in this comparison analysis Crash test FMVSS 208 NCAP Frontal IIHS Frontal FMVSS 214 NCAP Side IIHS Side Unique parameters Unbelted 5th percentile female and 50th percentile male, sometimes paediatric available, 40 kph (25 mph), full frontal Belted 5th percentile female and 50th percentile male, sometimes paediatric available, 56 kph (35 mph), full frontal Belted 50th percentile male driver only, 64 kph (40 mph), offset frontal Crabbed side impact, standard size impacting vehicle, mainly 50th percentile male, 53 kph (33 mph), near side Crabbed side impact, standard size impacting barrier, mainly 50th percentile male, 62 kph (39 mph), near side 90 side impact, large impacting barrier, 5th percentile female, 50 kph (31 mph), near side The next major parameters examined were vehicle make and model year, for which the sisters and clones list was employed. This list specifies which vehicles underwent significant safety modifications according to model year (sisters), and which others were sold under different brand names (clones) (Anderson, 2008). With this list it was possible to identify a range of vehicle model years that did not undergo major changes and to identify alternate vehicle brand names for which to search the crash test databases. Because this analysis focused on case-by-case comparisons, it was important to choose comparison crash test vehicles that were within the same generation as the CIREN case vehicle so that crashworthiness could be assessed. After setting search parameters based on the configuration of the CIREN crash and the make/model and year of the vehicle involved, typically only a few choices remained in the crash test database for comparison. If the initial database chosen did not contain a correct vehicle make/model, the alternate database was searched. This was typically enered with IIHS searches, because the IIHS tests fewer vehicles than the NHTSA. 2.2 Part 2: awarding similarity points After the most similar crash test was found for each CIREN real-world case, a quantitative method to identify the best comparison pairs was required. The SSM was developed to quantitatively identify the best pairs of comparisons between crash tests and the real-world CIREN cases. Each of the parameters used to identify comparisons in Part 1 was assigned either a 0 or 1 depending on the similarity to the crash test (Table 2). This resulted in a maximum score of 12. If a parameter was unknown for either the real-world case or the crash test, the comparison did not receive a similarity point for that parameter. For known parameters, similarity points were awarded if the following criteria were met. s were only scored if they were frontal crashes compared to frontal crash tests or side impacts in comparison to side impacts. All other crash types were excluded from analysis.

6 196 K.L. Loftis et al. Table 2 Crash, vehicle and occupant parameter requirements for similarity scoring between IIHS and NHTSA crash tests and CIREN real-world crashes Comparison characteristic Vehicle year Vehicle make/model PDOF DeltaV Crash type Impacted object Maximum crush Occupant position Belted status Airbag deployment Occupant height Occupant weight Requirement for point Sisters and clones list Sisters and clones list +/ 20 degrees +/ 16.1 kph (10 mph) Exact Similar shape/size +/ 10 cm (3.9 in) Exact for crash configuration Exact Exact for crash configuration +/ 7.6 cm (3 in) +/ 4.5 kg (10 lbs) There were seven crash and vehicle parameters included in the scoring system. Because there was a sequential methodology for selection of crash tests in comparison to the realworld crashes, the vehicle and crash parameters were given more emphasis and were more likely to be similar. 1 Vehicle year the case received a point if the case vehicle and crash test vehicle were within the model year range (Anderson, 2008). 2 Vehicle make/model the case received a point if the vehicles were the same or clone models (Anderson, 2008). 3 PDOF if the case vehicle PDOF was within 20 degrees of crash test PDOF, a point was received. 4 DeltaV as computed by WinSmash, the deltav had to be within 16.1 kph (10.0 mph) of the crash test deltav reported (Cheng et al., 2005). 5 Crash type the specific types of CIREN crashes were either full frontal or offset frontal, near side or far side, etc., and had to be the same as the crash test. No farside cases received a similarity point for this parameter, because such crash types were not available within the crash test databases. 6 Impacted object the size and shape of the impacted object had to be equivalent. For frontal impacts, either the CIREN case impacted another vehicle (in comparison with a frontal barrier crash test) or it impacted a pole/tree (in comparison with a frontal pole crash test). For side impacts, the CIREN case received one similarity point if it corresponded to the crash test with either passenger car impact, large vehicle impact or pole. 7 Maximum crush the case received a point if maximum crush was within 10.0 cm (3.9 in) of the maximum crush measurement reported by the crash test.

7 A similarity scoring technique 197 There were five occupant parameters included in the SSM. Occupant parameters are important when investigating occupant kinematics and injuries. As occupant seating position was a more important variable when selecting the initial crash test most similar to the real-world case, this parameter was more likely to be similar between the two cases. 1 Occupant position for frontal impacts, if the CIREN occupant seating position corresponded to the ATD seating position for driver or front passenger, or if the CIREN occupant and ATD were backseat passengers. For side impacts, CIREN occupant seating position had to correspond exactly with the seating position of the ATD. This was to ac for near- versus far-side occupants, and those either in the front or rear seat. As an example, if the CIREN occupant and ATD were both nearside occupants in the driver position for a side impact, this parameter received one similarity point. 2 Belted status belted status received a point if the real-world occupant belted status matched the crash test ATD belt status (belted, unbelted, child safety seat). 3 Airbag deployment to receive the similarity point, airbag deployment had to be the same as the crash test, according to the crash configuration. Only frontal airbag deployment was assessed for frontal impacts, and only side airbag deployment was assessed for side impacts. Air bag deployment was assessed for a specific occupant seating position rather than air bag deployment for the general vehicle. 4 Occupant height a point was received if occupant height was within 7.6 cm (3.0 in) of the ATD height. Occupant height and weight were used instead of BMI to examine each one separately, as each may play a role in injury mechanisms. The most common crash test ATD s and their height and weight information can be seen in Table 3. 5 Occupant weight Occupant weight received a point if it was within 4.5 kg (10.0 lbs) of the ATD weight. The weight for the ES-2re ATD was adjusted to match the weight of a 50th percentile male with arms. Therefore, the weight of the ES-2re is listed as that of the 50th percentile male at 78.0 kg (172 lbs) because all of the comparison real-world occupants weights included upper limbs. This same weight adjustment was used on the SID-IIs, because this ATD does not have a right arm. The weight used for the SID-IIs is the weight listed for the 5th percentile female. Table 3 Common ATD heights and weights used by NHTSA and IIHS Frontal Side ATD Height Weight 50th % male HIII 175 cm (69 in) 78 kg (172 lb) 5th % female HIII 150 cm (59 in) 49 kg (108 lb) 3 year old HIII 94 cm (37 in) 16 kg (35 lb) ES-2 (50th % male) 175 cm (69 in) 78 kg (172 lb) SID (50th % male) 175 cm (69 in) 77 kg (170 lb) SID-IIs (5th % emale) 150 cm (59 in) 49 kg (108 lb) The height difference range of 7.6 cm ( 3 in) to 7.6 cm (3 in) was selected to represent an overall spread of 15.2 cm (6.0 in). A smaller height range was too selective, but a

8 198 K.L. Loftis et al. larger range was not discriminating enough for assigning similarity points. A height difference range of ± 15.2 cm (6 in) would have resulted in 77 cases that received a similarity point with ATD height and would have allowed a 30.5 cm (1 ft) spread for the acceptable range. The height range was also selected to prevent overlap between ATD heights for similarity points (15 cm [10 in] difference between 50th percentile male ATD height and 5th percentile female ATD height). Occupant weight had the greatest variation of all parameters. To avoid allowing large weight ranges for small height ranges, the weight range was chosen based on the height range using body mass index (BMI) values. As an example, the 50th percentile male ATD has a BMI of Raising the height by 7.6 cm (3 inches) while keeping the same BMI would require a weight increase of 5.4 kg (12 lb). By setting the weight difference range within reasonable values for the height difference range, obese occupants could receive the height similarity point but not receive the weight similarity point in comparison to the ATD. 3 Results Of the 100 comparison cases, 66 were frontal impacts and 34 were side impacts. There were 60 passenger cars, 17 SUV s, 14 pickup trucks and 9 vans; model years ranged from 1997 to There were 63 females and 37 males. CIREN case occupants were most commonly drivers (73%) followed by front right passengers (19%) and rear seat passengers (8%). The average age of female case occupants was 47.5 years, and the average age of male case occupants was 40.8 years, resulting in an overall case occupant average age of 45.0 years. 3.1 Part 1: results from selecting similar crash test for each CIREN case After selecting the best crash test for each CIREN real-world case, results revealed that the NHTSA database was the source of 67% of the comparison crash tests; the IIHS database, 33%. Table 4 shows that IIHS frontal impacts (27 cases, 41% of frontal comparisons) and NHTSA side NCAP (18 cases, 53% of side comparisons) were the crash tests most frequently used for comparison. Of the 66 frontal-impact tests identified for comparison, 57 used 50th percentile male ATDs; 7 used 5th percentile female ATDs and 2 used a HIII 3-year-old ATD seated in a child safety seat (CSS). Of the 34 sideimpact crash tests chosen for comparison, 24 used SID ATDs; 8 used SID-IIs ATDs and 2 used ES-2 ATDs. Table 4 Type of crash test Frontal totals Counts and percentages of comparison crash tests for frontal and side impacts Count and percentage of each type 66 cases total NHTSA NCAP 26 (39%) NHTSA FMVSS 208 barrier or pole 13 (19%) IIHS frontal 27 (41%) Side totals 34 cases total NHTSA SNCAP 18 (53%) NHTSA FMVSS 214 barrier or pole 9 (26%) IIHS side 7 (21%)

9 A similarity scoring technique Part 2: similarity scoring results In analysis of the overall similarity scores resulting from the SSM, Figure 1 shows the distribution of similarity scores for the CIREN cases. Scores ranged from 4 to 11 points, with an average of 7.8 points. CIREN cases that received a similarity point for each parameter are shown in Figure 2. After the comparisons were scored, specific parameters were investigated. As this SSM used a sequential methodology, the most important parameters for comparison were investigated first. Parameters that were identified through expert opinion to have a greater effect on the vehicle dynamics and occupant kinematics were more often the parameters with higher similarity to the crash tests, because these were given greater significance in the crash test selection process. Figure 1 Similarity scoring results for all 100 comparison cases between CIREN real-world crashes and crash tests Figure 2 Number of CIREN cases receiving similarity points for each parameter in comparison to a crash test

10 200 K.L. Loftis et al Crash and vehicle characteristic similarity scoring The CIREN crash and vehicle parameters awarded the most similarity points were vehicle year (98) and vehicle make/model (92). The fewest similarity points were for deltav (39), the object impacted (56) and the maximum crush measurement (35). These three parameters were examined further. DeltaV deltav differences were computed by subtracting the real-world deltav from the crash test deltav. Ranges receiving the similarity point are designated with the patterned bars in Figure 3. Most real-world crashes had deltavs below crash test deltavs. Object impacted the impacted objects in frontal crashes differed in shape, mainly because of a larger number of pole (or tree) impacts in the real-world crashes. Any CIREN case that impacted another vehicle received one similarity point for object impacted, because of no alternate barrier sizes available in frontal crash tests. Frontal cases receiving one similarity point for the impacting object are listed in bold in Table 5. For side impact comparisons, there were multiple options in the crash test databases based on the size of the impacting object (Table 6). This difference in size of the impacting object, either a large impacting vehicle (IIHS) or a regular size impacting vehicle (NHTSA), created more opportunities for incorrect comparisons; and thus fewer similarity points were awarded for this parameter. Maximum crush there was a wide range seen in the maximum crush comparison ranges between crash tests and real-world crashes, with most crash tests having less maximum crush. The real-world crash maximum crush was subtracted from the crash test maximum crush to compute the difference. Crush differences that received a similarity point are designated with patterned bars in Figure 4. The difference in maximum crush as a result of difference in deltav was also analysed. This correlation has an R 2 value of as shown in Figure 5. Each point on the plot is one comparison. Figure 3 DeltaV comparison results between CIREN cases and crash tests, with patterned bars displaying the ranges that received a similarity point for this parameter. To calculate these ranges, the CIREN deltav was subtracted from the crash test deltav

11 A similarity scoring technique 201 Figure 4 Maximum crush comparison results between CIREN real-world crashes and crash tests, showing ranges that received similarity points with patterned bars. Ranges were determined by subtracting the CIREN maximum crush from the crash test maximum crush Figure 5 Difference between the CIREN real-world crash and crash test deltav versus the difference in maximum crush, where each point represents one comparison case

12 202 K.L. Loftis et al. Table 5 Frontal crash comparisons between CIREN real-world cases and crash tests for the parameter impacted objects. Numbers in bold represent comparison cases that received a similarity point for this parameter, while those numbers not bolded did not receive the similarity point CIREN Frontal impacts Crash test type Standard barrier Pole Standard vehicle 20 0 Large vehicle 16 0 Pole type 24 2 Other 4 0 Table 6 CIREN Side crash comparisons between CIREN real-world cases and crash tests for the parameter impacted objects. Numbers in bold represent comparison cases that received a similarity point for this parameter, while those numbers not bolded did not receive the similarity point Side impacts Crash test Standard barrier Large barrier Pole Standard vehicle Large vehicle Pole type Other Occupant characteristic similarity scoring The occupant parameters awarded the most similarity points for their similarity to crash tests were seating position (86) and airbag deployment (86). The fewest similarity points were belted status (73), occupant height (41) and occupant weight (20). Belted status most CIREN cases that did not receive a point for this parameter were those in which the real-world case occupant was unbelted and the ATD was belted (22 cases). Comparisons that received a similarity point for this parameter are shown in bold in Table 7. There were also four paediatric real-world case occupants that involved CSS usage. Two of these CIREN cases had comparison crash tests that used a paediatric ATD in CSS, and two had comparison crash tests that did not. The two cases that both used a CSS were awarded one similarity point for this parameter. Occupant height to investigate differences in height, the real-world occupant height was subtracted from the height of the ATD used for that particular crash test. In the comparisons, six different ATDs were used. The results of this height difference analysis revealed that many CIREN occupants are shorter than the 50th percentile male ATD (Table 8). Occupant weight to investigate differences in weight, the real-world occupant weight was subtracted from the weight listed for the ATD used for that particular crash test comparison. The differences in weight are shown in Table 9, with the bolded numbers representing the cases that received a similarity point. CIREN occupants were generally heavier than the ATDs used in the comparison pairs. The difference in occupant height compared with the difference in occupant weight was analysed, and the plot is shown in Figure 6, with an R 2 value of Each point on the plot represents one comparison.

13 A similarity scoring technique 203 Table 7 CIREN Table 8 Three-point lap and shoulder belt comparison results for adult occupants, with numbers in bold designating comparisons between CIREN cases and crash tests where a similarity point was received for occupant belted status Crash test Three-point lap and shoulder belt usage Belted Unbelted Belted 65 3 Unbelted 22 6 Height difference results where the CIREN occupant height was subtracted from the ATD height. Numbers in bold received the similarity point for this parameter because these were within 7.6 cm (3 in) of the ATD height Difference in height (DH) DH = ATD occupant HIII 50th % male HIII 5th % female 3 yr old ES-2 SID SID-IIs Total Occupant taller cm DH 7.6 cm ATD taller Total Table 9 Weight difference results between the CIREN occupant and the crash test ATD, where numbers in bold represent comparisons that received a similarity point because the CIREN occupant was within 4.5 kg (10 lbs) of the ATD weight Difference in weight (DW) DW= ATD - Occupant HIII 50th % male HIII 5th % female 3 yr old ES-2 SID SID-IIs Total Occupant heavier kg DW 4.5 kg ATD heavier Total Figure 6 Difference between CIREN occupant and crash test ATD occupant height versus difference in occupant weight, where each point represents one comparison case

14 204 K.L. Loftis et al. 4 Discussion This study applied 12 comparison parameters to 100 CIREN frontal and side impact cases and created the SSM to identify the best comparisons. Because of variations in crash, vehicle and occupant parameters, the SSM differentiated between real-world cases that were more or less similar to crash tests as shown by the range of resultant similarity scores (Figure 1). They were ranked having low (0 6 points, 18 cases), medium (7 9 points, 72 cases) or high (10 12 points, 10 cases) similarity. Future research involves using the cases that scored well ( high similarity) to investigate ATD injury risks in comparison with the real-world injuries. The results of this study also reveal areas (such as deltav or crash type) where a weighted scoring system may be useful to further differentiate comparisons that were highly similar between the crash test and real-world case. By examining each of the 12 parameters and the range of parameter values, areas of high similarity (such as crash type, airbag deployment) are shown, as well as areas of low similarity (deltav, maximum crush, occupant height and weight). Future iterations of the SSM can benefit from normalising the continuous parameters, where a comparison case could score anywhere between 0 and 1 based on the difference between the crash test value and the real-world value. 4.1 Vehicle and crash parameter similarity scoring Using the sisters and clones list allowed the vehicle year and make/model to compare well between the real-world crash and crash test (Anderson, 2008). These two parameters were two of the first selection criteria used in the sequential selection of a crash test to a real-world case. There were only two CIREN cases for which the vehicle year range was not available in either the NHTSA or IIHS databases, demonstrating that most vehicles are available in crash test databases. While there are vehicle mass differences between similar model years and vehicle make/models, these differences were thought to have little effect on the outcome of the crash test or real-world crash. If large vehicle mass differences were generated between vehicle make/model years, the safety systems in those vehicles would be adjusted to maintain crash test performance. This would lead to the vehicle no longer being a sister or clone. Future iterations of this scoring system could include a parameter for vehicle curb weight to further investigate these vehicle size differences. When selecting comparisons based upon crash configuration (frontal or side) there were a finite number of choices for PDOF differences, resulting in close comparisons for many CIREN cases. Because of the hierarchy comparison methodology where vehicle year/make/model and crash type was chosen first, secondary items that could not be controlled as closely did not compare as well, including deltav, impacted objects and maximum crush. For these parameters, there was a wide range of values for the realworld crash, but only a select range for the crash tests. Because crash tests are run at set speeds, into prescribed barriers, they result in standard crush measurements for each vehicle. Maximum crush, deltav and the impacting object are all closely tied when investigating crash severity. Maximum crush values revealed there were 26 CIREN cases with maximum crush less than the crash test and 27 with maximum crush greater, showing a wide range in real-world crush values (Figure 3). The real-world maximum crush values may have been closer to crash test values if the impacting objects had also been more similar to the

15 A similarity scoring technique 205 barriers used in crash tests. Real-world pole-type impacts often involve trees, which can yield during a crash, thus reducing the damage to the vehicle and reducing the maximum crush. Narrow rigid pole impacts have been shown to generate higher crush and intrusion measurements than distributed crashes (Hassan, 2005). These differences contributed to the varying maximum crush differences seen for the comparison cases. For side impacts, only near-side impacts were utilised. The kinematics between nearand far-side impacts are different and result in different injury patterns. In a far-side crash, the occupant can move out from under their seatbelt resulting in more excursion and injurious contacts. Currently, there are no standard far-side crash tests. For side impacts using the SSM, specific impact location differences on the side of the vehicle can be analysed on a case-by-case basis after the initial best comparison real-world cases are selected using the final scores. It is also suggested that future iterations of the comparison and similarity scoring algorithm may include specific information about the area of damage on the side of the vehicle, utilising the CDC code to identify whether the maximum crush was in the front, passenger compartment or rear of the side of the vehicle. For deltav, 37 real-world comparison cases had a lower deltav than the comparison crash test indicating that CIREN occupants were injured at speeds below the threshold tested by NHTSA and IIHS. Only six of the cases not receiving the similarity point for deltav had a higher deltav than the comparable crash test. Future iterations of this SSM could be developed to use the percentage difference in deltav for determining which cases receive a similarity point. This would allow similarity points for larger differences at higher deltavs but for smaller differences at lower deltavs. Because this SSM was the first implementation of a scoring algorithm and all the crash tests have moderate-to-high deltavs, comparing to low deltavs was not of particular concern here. DeltaV difference also correlated with maximum crush difference between the crash test and the real-world crash with a moderate R 2 value. A greater difference in deltav showed a greater difference in maximum crush. This result was expected, because CIREN deltav estimates were made using crush measurements in the WinSmash program. For the CIREN dataset used for the creation of the SSM, deltav estimation with WinSmash was completed using categorical stiffness values from size classes (pickup, van, etc) as determined from crash tests (Sharma et al., 2007, Wang and Gabler, 2007). There are known discrepancies within the WinSmash program that can create over/under estimations of deltav (Cheng et al., 2005, Niehoff and Gabler, 2006). Maximum crush, direction of impact and location of impact are input into WinSmash to calculate the estimated deltav. The impacting object parameter had more specific options for side impacts than for frontal impacts, which resulted in fewer similarity points. Before 2011, few crash tests with pole impacts or barriers of a different size for frontal impacts were available in the crash test databases. Because of this limitation, an impacted vehicle of any size in frontal impacts received a similarity point in comparison with a frontal barrier crash test. Realworld tree/pole impacts resulted in the largest percentage of cases that did not compare to crash tests for object impacted, because most of the crash tests were run with wide barriers. By incorporating side crash tests from the IIHS database, barrier size choice was varied for side impacts. CIREN cases received a similarity point if the crash test barrier was similar in size to the impacting vehicle size in the real-world side impact crash.

16 206 K.L. Loftis et al. Side impacts, where the impacting vehicle was any type of passenger car, received a point for comparison to NHTSA side crash tests. Side impacts where the impacting vehicle was an SUV, truck or van received a similarity point for comparison with an IIHS side impact with the larger barrier. Using the SSM, real-world cases similar to a crash test are identified for further research. Parameters not available in crash tests, such as vehicle size mismatch and over/underride in the real-world crash, can then be qualitatively investigated on a case-by-case basis as a possible reason the real-world case occupant sustained injury while the crash test passed injury risk thresholds. Using a combined quantitative and qualitative approach, these comparison techniques can be used to look at differences in occupant injuries due to side impact vehicle mismatch crashes, where the impacting vehicle is larger and higher than the struck vehicle (Loftis et al., 2009). 4.2 Occupant parameters Occupant position and airbag deployment compared most frequently between the realworld and crash test occupant parameters. Occupant position compared well because most of the CIREN case occupants were drivers and all standard crash tests had an ATD seated in the driver s seat. Airbag deployment compared well because of the vehicle model year restrictions for recruitment within CIREN. All vehicles in this range were equipped with frontal airbags, and only a few had side airbags at the time this study was performed. As future safety enhancements are made to vehicles, more differences will emerge in crash comparisons. Occupant belt status differed between the real-world crashes and crash tests because many CIREN occupants were unbelted (Table 7). For this study, the unbelted FMVSS 208 test was chosen as a possible comparison crash test because many occupants in CIREN crashes were unbelted. A few NHTSA crash tests were found with unbelted ATDs for FMVSS 208 to correctly compare cases (six cases), but there were 22 other real-world crashes with an unbelted occupant where the crash test included a belted ATD. CIREN is an injury database and therefore more unbelted occupants are expected in CIREN cases. The SSM demonstrates that it is important to continue to study unbelted occupants involved in MVCs because unbelted occupants generally have more severe injuries and a higher risk of mortality than belted occupants (Crandall et al., 2001). There were also four real-world cases involving paediatric case occupants in Child Safety Seats (CSS). Standard crash tests containing paediatric case occupants were not typical, but comparison cases with paediatric ATDs in CSS were found for two of these cases. One difference between the CIREN paediatric cases and the crash tests run with paediatric ATDs was the proper use of the CSS. This was ensured for the crash test, but many CIREN paediatric occupants may have been incorrectly restrained (Loftis et al., 2011a). While correct CSS use results in fewer injuries and deaths, incorrect use has been shown to result in unique injury patterns and more serious injuries for paediatric occupants (Tyroch et al., 2000; Brown et al., 2006; Bulger and Kaufman, 2006; Arbogast et al., 2009). By performing more of these comparisons, it will be possible to look at the differences in injury probability from ATD measurements and paediatric CIREN occupant injuries for improperly restrained occupants. Occupant height and weight did not compare well between CIREN cases and crash tests. Twenty-one real-world case occupants were taller than the comparable ATD, while 38 were shorter, as shown in Table 8. Half of the real-world case occupants weighed

17 A similarity scoring technique 207 more than the ATD selected for that comparison case. Only 30 ATDs weighed more than the case occupant, and all were the ATD representing the 50th percentile male, as shown in Table 9. Comparing occupant height difference versus occupant weight difference showed that greater occupant height difference resulted in greater weight difference; but this had a low R 2 value. Many of the real-world occupants were considered overweight or obese according to medical records. Recent studies have shown that obese occupants have different kinematics during a crash, even when belted, and increased body weight may increase injury severity for occupants in a crash (Viano et al., 2008; Forman et al., 2009; Urban et al., 2010). More research should be done to investigate the influence of occupant weight on injuries sustained and patient outcomes. Instead of using BMI, this similarity scoring analysed weight and height separately, allowing specific differences in height and weight to be examined. While obesity is determined using BMI, using BMI for similarity scoring would not have allowed comparison between occupants known to be shorter and lighter or taller and heavier than an ATD, as the BMI might have been the same. 4.3 Use of similarity scoring system and future work The comparison methodology is used to find the best crash test comparison for each CIREN case based on a hierarchy of important vehicle, crash and occupant parameters. The SSM is then used to determine which of these comparisons result in the most similar cases. The comparisons that are most similar (with highest scores) will be used in the future to analyse the specific differences that resulted in injury for the real-world case occupant while the crash test results did not exceed set thresholds for safety. To study specific differences, important parameters such as the 12 identified in the SSM need to be similar to assess injury outcomes. If many parameters are dissimilar, it is unknown which dissimilarity most likely led to injury for the case occupant. Using a comparison methodology and similarity scoring system such as the SSM, it is possible to find cases similar to one another (real-world cases to crash tests, or real-world cases to other realworld cases) to study the effects specific dissimilarities have on occupant safety and injury risk. These comparison and similarity scoring techniques are already being used in CIREN case review meetings to investigate vehicle dynamics and occupant kinematics during crashes (Stitzel et al., 2009; NHTSA, 2010a). After identifying the similar crash test for the CIREN case, crash test reports, acceleration pulses and videos are studied to help researchers determine vehicle dynamics and occupant kinematics for the real-world CIREN case. For the comparisons that receive high scores (i.e. most similar), the crash test ATD measurements and injury risks can elucidate how the real-world CIREN injuries occurred. Establishing baseline similarity scoring methods is a necessary first step before proceeding with injury studies, to ensure the cases are close comparisons for extracting injury conclusions. This study provides a standard, repeatable comparison and similarity scoring methodology for comparing real-world crashes to crash tests on a caseby-case basis. As an example, frontal crashes with high similarity between the real-world crash and crash test could be used to investigate occupant kinematics and knee-to-knee bolster contacts resulting in knee-thigh-hip fractures (Rupp et al., 2002; Rupp et al., 2010). Most crash test ATD measurements do not exceed injury risks during crash tests, but age and

18 208 K.L. Loftis et al. comorbidities are not currently factors in injury risk thresholds that determine vehicle crashworthiness. It is possible to investigate occupant parameters that may pre-dispose someone to a specific injury by studying real-world cases that are similar to a frontal crash test. Occupant comorbidities such as osteopenia, osteoporosis and surgical implants (knee or hip replacements) can make a real-world occupant more susceptible to injury. Comorbidities and occupant age are not currently taken into ac when analysing crash test results. With more research to quantify exactly how much these comorbidities (or other parameters) contribute to injuries in real-world crashes, they could be incorporated into revised injury risk curves to use with ATD measurement data to access new injury risks for high-risk motor vehicle occupants. This analysis is only possible when cases that are similar to crash tests are used to investigate outside parameters that may be affecting injury risk for occupants. The SSM can also be applied to comparisons between real-world cases, to compare safety systems between older and newer vehicles and crash tests, as shown in previous work (Stitzel et al., 2009). Real-world safety system comparisons are important for investigating different injury outcomes based on key dissimilar points. With side impact airbags becoming industry standards, there will likely be a decreased incidence of thorax and head injuries in near-side impacts in the future (Tencer et al., 2005; NHTSA, 2007). Occupant injury and safety system trends can be investigated not only using crash tests, but also in the comparative real-world crashes in the CIREN database (Loftis et al., 2011b). Evaluation of the methods using a population-based sample such as NASS-CDS could allow for better evaluation of the ranges of parameters which should be considered similar. This approach can be based on reasonable targets for the number of matching cases or particular output parameters for evaluating ATD injury risk estimates with realworld occupant injury incidence. A NASS-CDS study, especially, would aid in determining the ranges that receive a similarity point for the numerical parameters, such as deltav, maximum crush, occupant s height and weight. As this work was a first look at developing a similarity scoring methodology, it is expected that future iterations may investigate using normalising for continuous parameters and including a weighting scheme based on the hierarchy of crash parameters. There are also some crash types that are currently unavailable in both crash test databases, including far-side impacts and rollovers. Multiple studies have shown that occupant kinematics are different for a far-side occupant, and they can move out from under a seatbelt to contact a near-side occupant in a crash (Bostrom and Haland, 2003; Fildes et al., 2003). Comparisons between far-side real-world occupants and near-side ATD s may reveal specific differences that could be further investigated through the addition of a far-side ATD to the side impact crash tests. The results of this comparison study identify specific crash, vehicle and occupant parameters that may benefit from further vehicle safety enhancements or revised crash test requirements to mitigate serious injuries in the future. Incorporating new injury risk curves designed to estimate injury risk for overweight occupants is just one way these results could be used. Research emphasis on far-side occupants is also encouraged, as there are currently no standard far-side crash tests available. Despite all CIREN cases including injured occupants, crash deltavs and maximum crush were still lower for most cases than crash test values. This work demonstrates that while crash tests are run at extreme speeds and configurations, occupants are being injured in crashes below crash