Im

Aa

b

a

ARRA

KHFITRM

1

mtyd(

eT

kj

0h

Accident Analysis and Prevention 64 (2014) 1– 8

Contents lists available at ScienceDirect

Accident Analysis and Prevention

journa l h om epage: www.elsev ier .com/ locate /aap

njury prediction in a side impact crash using human bodyodel simulation

dam J. Golmana,b, Kerry A. Danelsona,b, Logan E. Millera,b, Joel D. Stitzela,b,∗

Virginia Tech-Wake Forest University Center for Injury Biomechanics, Medical Center Boulevard, Winston-Salem, NC 27157, USAWake Forest University School of Medicine, Medical Center Boulevard, Winston-Salem, NC 27157, USA

r t i c l e i n f o

rticle history:eceived 8 May 2013eceived in revised form 16 October 2013ccepted 23 October 2013

eywords:uman body modelinite element analysisnjury metricshoracic injuryeal worldotor vehicle crash

a b s t r a c t

Background: Improved understanding of the occupant loading conditions in real world crashes is criticalfor injury prevention and new vehicle design. The purpose of this study was to develop a robust method-ology to reconstruct injuries sustained in real world crashes using vehicle and human body finite elementmodels.Methods: A real world near-side impact crash was selected from the Crash Injury Research and Engineer-ing Network (CIREN) database. An average sedan was struck at approximately the B-pillar with a 290degree principal direction of force by a lightweight pickup truck, resulting in a maximum crush of 45 cmand a crash reconstruction derived Delta-V of 28 kph. The belted 73-year-old midsized female driver sus-tained severe thoracic injuries, serious brain injuries, moderate abdominal injuries, and no pelvic injury.Vehicle finite element models were selected to reconstruct the crash. The bullet vehicle parameters wereheuristically optimized to match the crush profile of the simulated struck vehicle and the case vehicle.The Total Human Model for Safety (THUMS) midsized male finite element model of the human body wasused to represent the case occupant and reconstruct her injuries using the head injury criterion (HIC),half deflection, thoracic trauma index (TTI), and pelvic force to predict injury risk. A variation study wasconducted to evaluate the robustness of the injury predictions by varying the bullet vehicle parameters.Results: The THUMS thoracic injury metrics resulted in a calculated risk exceeding 90% for AIS3+ injuriesand 70% risk of AIS4+ injuries, consistent with her thoracic injury outcome. The THUMS model predictedseven rib fractures compared to the case occupant’s 11 rib fractures, which are both AIS3 injuries. The

pelvic injury risk for AIS2+ and AIS3+ injuries were 37% and 2.6%, respectively, consistent with the absenceof pelvic injury. The THUMS injury prediction metrics were most sensitive to bullet vehicle location. Themaximum 95% confidence interval width for the mean injury metrics was only 5% demonstrating highconfidence in the THUMS injury prediction.Conclusions: This study demonstrates a variation study methodology in which human body models canbe reliably used to robustly predict injury probability consistent with real world crash injury outcome.

. Introduction

Motor vehicle crashes (MVCs) account for approximately 1.2illion deaths each year worldwide and are predicted to become

he fifth leading cause of death by 2030 (WHO, 2009). Each

ear in the United States, MVCs are responsible for 30 thousandeaths, 1.5 million injuries, and over $70 billion of lifetime costsNaumann et al., 2010). Improving the understanding of injury

∗ Corresponding author at: VT/WFU School of Biomedical Engineering and Sci-nces, 575 N. Patterson Avenue, Suite 120 Winston-Salem, NC 27101, USA.el.: +1 336 716 5597; fax: +1 336 716 5491.

E-mail addresses: adam.golman@gmail.com (A.J. Golman),danelso@wakehealth.edu (K.A. Danelson), loganem8@vt.edu (L.E. Miller),stitzel@wakehealth.edu, jdstitzel@gmail.com (J.D. Stitzel).

001-4575/$ – see front matter © 2013 Elsevier Ltd. All rights reserved.ttp://dx.doi.org/10.1016/j.aap.2013.10.026

© 2013 Elsevier Ltd. All rights reserved.

mechanisms and the effectiveness of injury mitigation systemsin real world crash scenarios is important for the design of safervehicles (Yoganandan et al., 2007). For this reason, human bodyfinite element models (FEMs) have been developed to investigateinjury and crashworthiness of motor vehicles at a level of detaildifficult to achieve with physical tests with anthropomorphic testdevices (ATDs).

The field of injury biomechanics has a long history of usingnumerical models of the human body (Yang et al., 2006). Previousstudies have used FEMs of isolated anatomical regions includingthe brain (Zhang et al., 2001; Takhounts et al., 2008), eye (Stitzelet al., 2002), aorta (Shah et al., 2001), and lungs (Gayzik et al.,

2007), to develop novel organ level injury metrics. When investi-gating impact scenarios involving interactions with the entire body,human body models (HBMs) are a valuable tool (Gayzik et al., 2012;Vavalle et al., 2012). An example of a HBM is the Total Human Model

2 nalysis and Prevention 64 (2014) 1– 8

fUfiesicfemiassH

afifri(tcreM

2

uCtemHsPW

2

dcroa(sfT1ee

tAsiHiqm

Table 1Reconstructed case occupant AIS2+ injuries.

AISCODE Description

442203.4 Left pneumothorax441408.3 Left upper and lower lobe contusions450203.3 Left 1st thru 10th; Right 3rd rib fracture140651.3 Left frontal subdural hemorrhage (SDH)140694.2 Left temporoparietal subarachnoid hemorrhage

(SAH) in the cortical sulci

Table 2Additional case occupant AIS2 injuries not compared to simulation results.

AISCODE Description

544222.2 Grade 1 splenic injury with small amount of fluid at theinferior pole

541610.2 Left kidney contusion840405.2 Minimally displaced medial tibial plateau avulsion fracture854463.2 Right bimalleolar fracture650220.2 Left C7 transverse process fracture

Table 3Validation tests conducted by NCAC for the updated Taurus model using NationalHighway Traffic Safety Administration (NHSTA) and Insurance Institute for HighwaySafety (IIHS) crash test data. MDB = moving deformable barrier.

Crash test type Comparison crash test data

Frontal, full wall (56 kph) NHTSA: 3248, 4150, 4776, 5143Frontal, full wall (48 kph) NHTSA: 3150, 3224, 4134, 4135, 4174Frontal, offset (64 kph) NHTSA: 3365

Tacoma frontal stiffness was calculated from the force-stroke datafrom full frontal crash tests which indicated that the RAV4 wasa valid substitute for the Tacoma up until 15 cm of stroke. TheTacoma average frontal stiffness was estimated from NHTSA test

Table 4Tacoma to RAV4 vehicle comparison.

Comparison parameter 1997 Toyota Tacoma 1997 Toyota RAV4

Weight (kg) 1245 1265Front overhang (cm) 72 74

A.J. Golman et al. / Accident A

or Safety (THUMS) (Iwamoto et al., 2002; Shigeta et al., 2009).sing THUMS, the injury potential of different astronaut suit con-gurations in various loading conditions was evaluated (Danelsont al., 2011a; Golman et al., 2012). THUMS has also been used totudy the effects of seatbelt location on bilateral carotid arterynjuries in far side impact (Danelson et al., 2009). The biomechani-al response differences in a Hybrid III and THUMS was investigatedor various belt and airbag configurations in a frontal crash (Mrozt al., 2010). One of the advantages of using a HBM is that the kine-atics, rib strains, and internal organ pressures can be analyzed to

nvestigate the effectiveness of injury mitigation systems (e.g. sideirbags) (Hayashi et al., 2008). All of these aforementioned HBMtudies used THUMS to investigate the human body response inimulated controlled laboratory tests. Another application of usingBMs is to reconstruct real world crashes.

HBMs can be used to predict injury risk from real world crashes, desirable capability for improving automobile safety. One of therst versions of THUMS was used to successfully reconstruct bone

racture in a real world frontal crash (Iwamoto et al., 2002). Aorticupture modes were investigated by reconstructing real world sidempacts from the Crash Injury Research and Engineering NetworkCIREN) database (Siegel et al., 2010; Belwadi et al., 2012). Whilehese studies reconstructed real world crashes, no study hasomprehensively analyzed the HBM response and predicted injuryisks. The purpose of this study was to build the framework andvaluate the THUMS capability to predict injuries in a real worldVC.

. Methods

A real world crash from the CIREN database was reconstructedsing complete vehicle finite element models from the Nationalrash Analysis Center publically available database and a state-of-he-art human body model. Simulations were performed using thexplicit finite element software, LS-DYNA MPP971 R4.2.1 (Liver-ore Software Technology Corporation, Livermore, CA), on a Redat Linux computational cluster. All output data from the LS-DYNA

imulation was automatically batch processed using the Injuryrediction Post Processor (IPPP), custom in-house MATLAB (Math-orks, Natick, MA) software.

.1. The real world crash

The real world side impact crash was selected from the CIRENatabase based on its similarities to the Federal Motor Vehi-le Safety Standard (FMVSS) 214 and common injury frequencyeported in side impact crashes. The case occupant was a 73-year-ld female with 173 cm height and 75 kg weight; therefore, hernthropometry approximately represented a 50th percentile male175 cm and 77 kg). She was the belted driver of a 2001 Ford Taurusedan with no side air bag. The case vehicle was struck while per-orming a left hand turn through an intersection by a 1997 Toyotaacoma pickup truck resulting in moderate left side damage (CDC0LZEW3), a maximum crush of 44 cm at the B-Pillar, 11.7 kph lat-ral Delta-V and 20.1 kph longitudinal Delta-V as measured by thevent data recorder (EDR), and WINSMASH Delta-V of 28 kph.

The case occupant was presumed to be seated in an upright pos-ure with the seat adjusted to a mid-track position. She sustainedIS2+ injuries to her left lung, ribs, and brain that were recon-tructed through the simulations (Table 1). She also sustained othernjuries that were not compared to the simulation results (Table 2).

er thoracic and abdominal injuries were caused by the intrud-

ng left front door with documented contacts on the upper dooruadrants. The age of the occupant and her fragility was deter-ined to be a contributing factor to the severity of rib fractures

Rigid pole, offset (56 kph) IIHS: CF05001Rigid pole, offset (64 kph) IIHS: CF05002Side impact, MDB (62 kph) NHTSA: 3263

(Kent et al., 2009). Her head contact to the roof rail was inferred,as there was no contact evidence. The case occupant had no AIS2+pelvic injury, but large amounts of bruising due to a partial glutealmuscle avulsion. Using semi-automatic medical image segmenta-tion techniques (Danelson et al., 2007, 2011b; Urban et al., 2012)0.52% of the brain, 19.9% of the lung, and 0.15% of the spleen wereidentified as volumes indicative of the location and severity ofinjury.

2.2. Vehicle finite element models

Vehicle FEMs were chosen from the National Crash AnalysisCenter (NCAC) publically available database. The 2001 Ford TaurusFEM, which was the same model and year as the case vehicle, wasselected to represent the struck vehicle. The updated Taurus FEMhas additional validation from a side impact crash test as wellas a wider variety of frontal crash tests (NCAC, 2012) (Table 3).The bullet vehicle in the CIREN case was a 1997 Toyota Tacoma;however, the Tacoma FEM was not available in the NCAC database.The most similar vehicle FEM was a 1997 Toyota RAV4 based onthe vehicle weight and outer dimensions (Table 4). The RAV4 and

Front track width (cm) 143 148Overall width (cm) 169 170Bumper height (cm) 46.5 47.7Frontal stiffness (kN) 1360 1600

A.J. Golman et al. / Accident Analysis and Prevention 64 (2014) 1– 8 3

Table 5Input parameters varied for crush correlation optimization.

Input parameters Variable Min value Max value

Bullet longitudinal velocity (m/s) LongV 13.0 16.0Bullet lateral velocity (m/s) LatV 0.0 3.5

nwRt

2

wpTtmrtismtcbpir

2

pntmtmct

Ft

Bullet rotational velocity (rad/s) RotV 0.0 6.0Principal direction of force (degrees) PDOF 280 300Bullet location (cm) BLoc 250 313

umber 2992, 3673, and 4478 and RAV4 average frontal stiffnessas estimated from NHTSA test numbers 2496 and 3613. TheAV4 FEM was validated using frontal NCAP test number 2496 ashe only crash test comparison (NCAC, 2008).

.2.1. Heuristic optimization to match crush profileThe difference between the simulated and case vehicle crushes

as heuristically optimized to determine the vehicle FEMs’ initialositions and velocities (Table 5) over 115 simulations (Fig. 1).he initial orientation of the two vehicles was estimated usinghe PDOF and the Delta-V from the CIREN case data. To imple-

ent a lateral velocity (LatV) component for the RAV4, friction wasemoved between the RAV4 tires and the ground. The bullet loca-ion (BLoc) was measured from the front of the Taurus to the initialmpact point on the RAV4 bumper. The simulated crush points wereelected to match the approximate location of the crush measure-ents as the CIREN case. The sum squared error (SSE) between

he simulation and case was calculated for each vehicle to vehi-le simulation. SSE was calculated for the front portion (C6, C5, C4),ack portion (Maximum crush point, C3, C2, C1), and the total crushrofile. Once the crush profile match was optimized, the HBM was

nserted into the vehicle and the HBM injury metrics and predictedisk were compared to the occupant injuries.

.3. Human body model

The case occupant was represented by THUMS version 4, 50thercentile male (178.6 cm and 74.3 kg), a validated HBM withearly 2 million elements representing the complex geometry ofhe human body (Shigeta et al., 2009). The differences in stature and

ass between the model and actual case occupant were assumed

o be negligible. THUMS was gravity settled with the seat in the

id-track position in the ideal driving position based on the CIRENase report of the occupant (Fig. 2). The belt system was modeledo reflect a generic belt system. Generic belt material properties

ig. 1. Input parameters for crush correlation optimization. Note that the PDOF inhis figure is 290 degrees.

Fig. 2. THUMS seated in the driver seat illustrating the location of the occupantrelative to the B-pillar.

provided by a restraint manufacturer were used for the seat beltelements and belt webbing. A retractor with a 3 kN load limit wasimplemented at the D-ring. A slipring with a friction coefficientof 0.2 was defined at the buckle location. The buckle pretensionerpresent in the vehicle did not fire during the impact and was there-fore not simulated.

2.3.1. THUMS injury metricsThe THUMS response was evaluated using several injury metrics

implemented to mirror the instrumentation used in regulatorycrash tests (Kuppa et al., 2003) and validated for THUMS in sideimpact configurations (Golman et al., 2013). Head, rib, spinal, andpelvic accelerations were measured at the center of gravity of theanatomical region of interest using the constrained interpolationmethod described by Golman et al. (2013). Head acceleration wasused to calculate the head injury criterion (HIC36) (Kuppa, 2006).Left rib 4, rib 8, and the 12th thoracic vertebral body accelerationswere used to calculate the thoracic trauma index (TTI) (Eppingeret al., 1984; Morgan et al., 1986; NHTSA, 2011). Virtual chestbands(Golman et al., 2013; Hayes et al., 2013) were used to determinemaximum half deflection by measuring the maximum change indistance between left side and the centerline (sternum to spine)of THUMS (Kuppa et al., 2003). Forces from the nodes on the lat-eral portion of the pelvis were extracted to measure the forceapplied directly to the pelvis (Appendix A). These injury metricswere used to calculate injury risk using risk functions found in theliterature (Kuppa et al., 2003). Pelvic force was converted to pelvicimpactor force by multiplying by a numerically derived and vali-dated scale factor of 1.859 to calculate injury risk based on PMHSpelvic impactor based risk functions (Appendix A). Injury was con-sidered to occur when the injury risk exceeded the injury risk usedto set the Injury Assessment Reference Value (IARV) in the FMVSS214: IARV established head AIS2+ risk threshold for HIC36 was 50%,thoracic AIS3+ risk threshold for half deflection and TTI was 50%,and pelvic AIS3+ pelvic force risk threshold was 25% (Kuppa, 2006).Rib fractures were simulated using the element deletion method(Li et al., 2010).

A technique was developed and employed to implement age-based changes in the rib material model. In the past, rib materialproperties of human body models have been aged based on thefemoral cortical data due to limited data on the ribs (Kent et al.,2005; El-Jawahri et al., 2010). However, by combining the data setsfrom two previous studies, a statistically significant (p < 0.0001) lin-ear correlation between age and ultimate strain was constructed(r2 = 0.30) (Kemper et al., 2005, 2007). The combined Kemper et al.(2005, 2007) rib cortical bone ultimate strain and age regressionwas similar to previously established femur cortical bone ultimatestrain and age regressions (Burstein et al., 1976; Mccalden et al.,1993), especially after considering the different body regions tested(Fig. 3). This function was transformed from ultimate strain to plas-tic strain by subtracting yield strain (0.88% per Kemper et al., 2005).

Using Eq. (1) an ultimate plastic strain of 0.88% was used to repre-sent a 75 year old occupant.

ultimate plastic strain [unitless] = −383 age [years] + 37, 514

106

4 A.J. Golman et al. / Accident Analysis and Prevention 64 (2014) 1– 8

Ff

2

CesiftotBqtit

3

3

3

bcaatuw

feCVTim

3

ld

TF

Fig. 4. Final vehicle crush comparison between case (black) and simulation (gold).The difference (cm) is shown between each crush point. (For interpretation of the

THUMS pelvic force of 4.3 kN (adjusted pelvic force of 7.9 kN),resulted in AIS2+ and AIS3+ pelvic injury risks of 37% and 2.6%,respectively.

ig. 3. Comparison between age effects on ultimate failure strain using regressionsrom Burstein et al. (1976), Kemper et al. (2005, 2007), and Mccalden et al. (1993).

(1)

.3.2. Variation studyDue to the many assumptions involved in reconstructing a single

IREN case (e.g. boundary conditions, vehicle and occupant mod-ls), a variation study was performed. The objective of this variationtudy was to quantify the sensitivity of the THUMS to variationsn crash parameters and develop a 95% confidence interval (CI)or the THUMS response for this CIREN case. In order to achievehese objectives, the bullet vehicle input parameters were variedne-factor-at-a-time (OFAT) from the baseline value to poten-ial alternative values specific to the CIREN case (See Appendix

for Table B1). The sensitivity of each variation parameter wasuantified using the coefficient of variation (CV) in each varia-ion parameter group (LongV, BLoc, etc.). The mean and CI of eachnjury metric and injury risk was determined using all 35 simula-ions.

. Results

.1. Baseline simulation

.1.1. Vehicle reconstructionFollowing the heuristic optimization, the final simulation

oundary conditions that best matched the simulated and caserush profile are listed in Table 6. The final simulated crush was

close match to the case crush (Fig. 4). The simulated crush had frontal, back, and total SSE of 49, 253, 302 cm2, respectively andhe crush area was 16.9% less than the case. Most notably, the sim-lation crush at C4 (the location of the occupant in the case) wasithin 2.24 cm of the real-world case.

The simulated vehicle CG delta-V was compared to EDR datarom the CIREN case vehicle. The 22.1 kph simulated vehicle CG lat-ral delta-V was 10 kph greater than the case EDR reported vehicleG delta-V, and the simulated 1.8 kph vehicle CG longitudinal delta-

was 11.4 kph less than the case EDR reported vehicle CG delta-V.he simulated lateral delta-V closely matched the case delta-V dur-ng the first 50 ms, which is the time frame when the peak injury

etric values occurred (Fig. 5).

.1.2. Injury predictionUnlike in the CIREN case, no head contact occurred for the base-

ine simulation. Similar to the CIREN case, the simulated intrudingoor contacted the THUMS thorax resulting in high risk values

able 6inal simulation boundary conditions.

Parameter LongV(m/s)

RotV(rad/s)

PDOF(degrees)

LatV (m/s) BLoc (cm)

Value 15 4 290 1.5 264.2

references to colour in this figure legend, the reader is referred to the web versionof this article.)

for thoracic injury. The door caused high lateral accelerations torib 4 (70 g), rib 8 (85 g), and T12 (99 g) resulting in a TTI of 200 g.TTI predicted a 92% risk of AIS 3+ injuries and 72% risk of AIS4+thoracic injuries. Using the virtual chest bands, the contours atthree different heights on the thorax were analyzed. The aver-age maximum half deflection was calculated for the upper (21%),middle (28%), and lower (26%) chest bands to assess injury risk.Using the maximum average maximum deflection (28%) a 96%risk of AIS 3+ injuries and 75% risk of AIS4+ injuries was pre-dicted.

Seven rib fractures were predicted compared to the case occu-pant’s 11 (Fig. 6), which are both AIS 3 injuries. While the simulationdid not predict all of the case occupant’s rib fractures, it did predictfour on the left side for ribs 7–10 and three on the right side for ribs4–6. The simulation under predicted fracture numbers on the leftside of the chest and over predicted fracture numbers on the rightside.

The simulation results predicted a low risk of pelvic injury keep-ing with the absence of pelvic injury in the case occupant. The

Fig. 5. Lateral vehicle CG delta-V comparison.

A.J. Golman et al. / Accident Analysis and Prevention 64 (2014) 1– 8 5

Fig. 6. Case occupant rib fractures (highlighted with red boxes) from CT reconstruc-tion (left) and predicted rib fractures from baseline simulation (right). Note that caseoccupant right rib 3 fracture is not shown. (For interpretation of the references tocolour in this figure legend, the reader is referred to the web version of this article.)

33veeh(PRba

3TtmwAf

va

TM

Fig. 7. Injury metric sensitivity measured by CV (%).

.2. Variation study

.2.1. Sensitivity. The sensitivity of each injury metric to eachariation parameter was quantified (Fig. 7). BLoc had the greatestffect on all injury metrics. The order of the remaining four param-ters was dependent on the specific injury metric. For TTI, LongVad the next greatest effect, while the effects of the other threeRotV, PDOF, LatV) were approximately equal. For half deflection,DOF had the next greatest effect, followed by LongV, LatV, andotV. For pelvic force, LatV had the next greatest effect followedy PDOF, RotV, and LongV. The injury risks follow the same trendss the input injury metrics.

.2.2. Distribution and CI for injury risks and critical injury metrics.he results from the variation study indicate a high risk of AIS 3+horacic injury, a medium risk for AIS 4+ thoracic injury, and a low to

edium risk of AIS 2+ pelvic injury. These injury risks compare wellith the injuries sustained by the case occupant, because she hadIS 3 rib fractures, an AIS 4 lung injury, and no pelvic injury except

or bruising of the hip and thigh from the pelvic impact (AIS 1).The CIs for the mean injury risk and critical injury metrics were

ery low, except for pelvic AIS3+ injury risk (Table 7). The meannd median for all CI values was 9% and 5% respectively. Despite a

able 7ean and 95% confidence interval for injury risks and critical injury metrics.

Metric Baseline Mean CI (%)

TTI (G) 200 182 3Thoracic AIS3+ risk (%): TTI 92 80 6Thoracic AIS4+ risk (%): TTI 72 51 12Half deflection (%) 28 27 3Thoracic AIS3+ risk (%): half deflection 96 90 5Thoracic AIS4+ risk (%): half deflection 75 63 10Pelvic force (kN) 4.3 4.6 5Pelvic AIS2+ risk (%): pelvic force 37 46 14Pelvic AIS3+ risk (%): pelvic force 2.6 4 38

Fig. 8. Rib fracture frequency over 35 simulations.

low CI for pelvic force, the pelvic AIS3+ risk CI was large due to thesteep slope of the injury risk curve at these force values. Therefore,small changes in force resulted in large changes in risk.

3.2.3. Rib fractures. All but one simulation (the most rearwardBLoc) predicted rib fractures with an average of 6 fractures. Mostfrequently, the fractures were posterior and to ribs 7, 8, and 9(Fig. 8). The most severe rib fractures in the case occupant wereribs 6, 7, and 8, which were in a similar location to the most com-monly fractured ribs in THUMS. In THUMS, left rib 9 fractured mostfrequently and was also most susceptible to a segmental fracture(posterior and lateral). Upon simulation visualization, the strainenergy typically propagated from the lower left ribcage though thebody and to the upper right which often resulted in right rib 4 andrib 5 fractures. Right side rib fractures only occurred after fracturesoccurred to the left ribs.

4. Discussion

This study used THUMS to predict the injuries in a real worldCIREN crash. All occupant injuries were successfully predictedexcept for head injury. To quantify the uncertainty in the injuryprediction, an OFAT variation study was performed. The resultingtight confidence intervals from this variation study suggested thatthe injury prediction values were both accurate and precise.

The uncertainty in the real world crash estimates was accountedfor by simulating variations of the CIREN case. Using CIs, the effectsof this uncertainty were bounded. The resulting small CI widthdemonstrates that a high level of confidence can be placed in theTHUMS risk estimates associated with the CIREN case. This varia-tion quantification based approach to crash simulation was a morecomprehensive method of crash assessment when compared tosimply reporting a single measurement. For this case, all but one CIwas between 3% and 14% of its mean value.

The injury metrics were most sensitive to changes in bulletlocation, because bullet location has the strongest influence onthe energy transmission directed into the occupant compartment,which was a main cause for injury. The injury metric sensitivity tothe other three parameters (RotV, PDOF, LatV) were all similar andchallenging to explain due to interaction effects not accounted forby the nature of an OFAT variation study. This limitation providesmotivation for performing a design of experiment (DOE) variationstudy in the future.

Head contact did not occur in either the baseline or variationstudy simulations. The lack of head contact during the simula-tions was most likely due to the lack of longitudinal delta-V ofthe struck vehicle. In the CIREN case, the vehicle experienced a12.5 mph longitudinal delta-V and 7.27 mph lateral delta-V. How-ever, in the reconstructions, there was a low longitudinal delta-Vcomponent (<4 mph). This longitudinal delta-V in the case causedforward excursion, enough to cause a hematoma on the right breastand bruising across the abdomen from belt loading. This forward

excursion may have placed the head more forward and at the sameheight as the roof rail and A-Pillar junction. Because the simulationscreated a low longitudinal delta-V, THUMS did not undergo sub-stantial forward excursion, and therefore the head did not impact

6 nalys

tiptHct

wsbbrdrrapTo

casfwmedmlfrmiao

emfopo2bbi

itwroco

5

cwTcml

A.J. Golman et al. / Accident A

he lower portion of the roof rail. Another possibility for no headmpact may have been due to the assumed position of the occu-ant; the occupant may have been leaning more forward duringhe crash event in reaction to the sight of an oncoming vehicle.owever because head contact was inferred (no evidence for headontact existed) in the CIREN case, the simulation comparison forhe head injuries were more difficult.

This study focused on predicting thoracic AIS2+ injuries, as theseere the most severe injuries sustained by the occupant. During

ide impact, the ribs play an important role in load transmissionetween the door and the occupant’s torso. To narrow the num-er of variables in our study, only the material properties of theibs were changed to reflect the occupant’s age and any age relatedegeneration of the bone. While only the material properties of theibs were changed to reflect age of the occupant, the nominal mate-ial properties implemented into THUMS are for about a middlege to elderly person. This is largely because the THUMS materialroperties tests were conducted with mostly elderly specimens.herefore, the overall age of THUMS is more representative of anlder person.

While gender differences in vulnerability in motor vehiclerashes do exists (Bose et al., 2011), these differences are typicallyttributed to females being smaller and lighter than males. Thisource of variation was one of the reasons this CIREN case chosenor this study, because the case occupant matched the height andeight of THUMS. Another potential reason for why females may beore vulnerable occupants is due to differences in material prop-

rties. However, there are a limited number of studies that haveirectly quantified the difference in material properties betweenales and females for various anatomical regions under impact

oading. The Kemper et al. (2005, 2007) data was used to accountor age related effects in the rib material properties, but not genderelated effects due to the small sample size. Despite using a maleodel, the predicted injuries were consistent with the occupant’s

njuries. Future studies could vary the rib material properties over range of values to understand the effects of both gender and agen injury risk.

The occupant’s anticipation of the crash and muscle activity lev-ls were not available in the CIREN report. The current THUMSodel does not have the capability to simulate bracing; there-

ore, these simulations represent a relaxed occupant. Even if theccupant was braced, the effects of bracing on occupant injuryotential are unknown for side impacts. Most studies have focusedn the effects of muscle bracing in frontal crashes (Beeman et al.,011; Iwamoto et al., 2012; Chang et al., 2009). Although theiomechanical effects of muscle bracing in side impacts have noteen studied, they warrant further investigation in future stud-

es.The occupant was assumed to be in a neutral upright position

mmediately prior to the crash. This assumption was supported byhe contact evidence on the driver door in the CIREN report. Thereere significant markings noted as denting and scratches in the

ear upper and lower quadrants of the driver door suggesting theccupant was in a normal driving position pre-crash. Future workould quantify the effects of seat position and posture on injuryutcome.

. Conclusions

This study demonstrates human body models such as THUMSan reliably be used to robustly predict injuries sustained in realorld crashes. A real world CIREN crash was reconstructed using

HUMS and vehicle finite element models. Besides the lack of headontact, the occupant kinematics and predicted injuries accuratelyatched the CIREN case. Following the completion of the base-

ine reconstruction, an OFAT variation study was performed to

is and Prevention 64 (2014) 1– 8

understand the parameter effect sizes and to bound the uncertaintyin the crash parameters associated with real world crashes. Themean injury risks and small confidence interval widths from thisvariation study demonstrated that the THUMS injury predictionswere precise. While this variation study was useful to understandthe parameter effect size on injury prediction, a larger DOEvariation study is necessary to more fully quantify correlationsbetween the parameters and the injury predictors. Future workcould simulate a wide range of side impacts to further understandcrash parameter effects, seat position, and age on injury metrics,organ strain metrics, rib fractures, and the effectiveness of injurymitigation systems.

Acknowledgments

Funding for this project has been provided by the ToyotaCollaborative Safety Research Center. Computations were per-formed on the Wake Forest University DEAC Cluster, a centrallymanaged resource with support provided in part by the Uni-versity. The authors acknowledge Mr. James Gaewsky for hisassistance in developing the THUMS pelvic force injury met-ric. Thanks to the National Crash Analysis Center at GeorgeWashington University for providing technical support for theirmodels.

Appendix A: Pelvic load cell implementation.

A pelvic load cell was implemented into THUMS in orderto predict pelvic injury. Injury risk functions were previouslydeveloped using impactor forces applied externally to the PMHSpelvis in various experimental tests. In these experimental tests,the total pelvic impact force for the PMHS was measured usingeither a load cell on a beam (Cavanaugh et al., 1990) or onthe head of a pendulum (Bouquet et al., 1998). Similar tothe development of the EuroSID2-re pubic symphysis pelvicinjury metric (Kuppa, 2006), a relationship between the impactorforce and THUMS pelvic force measurements must be estab-lished.

Before this relationship was established, the THUMS pelvicresponse was first validated in simulations mirroring the experi-mental set up of Cavanaugh et al. (side impact sled) and Bouquetet al. (pelvic block impact). In the side impact sled test, an ini-tial 6.7 m/s lateral velocity was applied to THUMS causing THUMSto collide with the rigid beams (Fig. A1a) (Golman et al., 2013),which included a pelvic beam that partially covers the pelvis(Fig. A1b). For the pelvic block impact, a 16 kg block, which fullycovers the pelvis, impacted the hip with a 10 m/s velocity (Fig. A1c),(Fig. A1d). Force transducers (RCForc) were used to measure forcesapplied to the impactor. The THUMS pelvic force measurementswere implemented using a nodal force group on the lateral sideof the pelvis, including the lateral surface of the ilium, ischium,and acetabulum, to measure forces transmitted directly to thepelvis.

Once the two simulations were validated (Figs. A2 and A3), theimpact velocities were varied to find a valid relationship betweenthe force exerted on the rigid impactor and the force transmit-ted to the pelvis across a broad energy spectrum. The side sledsimulation was performed with eight different velocities rangingfrom 4.7 to 9.48 m/s. The pelvic block simulation was performedwith nine different impactor velocities ranging from 7.1 m/s to14.1 m/s.

A linear regression was performed on the simulation resultsto determine the ratio of impactor force to the resultant pelvicforce. Despite different impactor geometries and regions of thepelvis impacted, these two tests produced similar ratios (Table A1).

A.J. Golman et al. / Accident Analysis and Prevention 64 (2014) 1– 8 7

Fig. A1. Cavanaugh et al. (1990) side impact sled test mirrored simulation (top) and Bouquet et al. (1998) pelvis block impactor mirrored simulation (bottom). (a) THUMScontacting the impactor and (b) close up of the impacted region of the pelvis. (c) Block impacting THUMS pelvis and (d) close up of the impacted region of the pelvis.

Fig. A2. Pelvic impact forces reported in the Cavanaugh et al. (1990) corridors.

Fig. A3. Pelvic impact forces reported in the Bouquet et al. (1998) corridors.

Table A1Summary of pelvic force and impactor/block force linear regression.

Ratio (pelvicforce/impactor force)

Coefficient ofdetermination (R2)

Tt

A

Table B1Variation study test matrix. Note that simulations are grouped by varied inputparameter.

Simno.

Sim name LongV(m/s)

RotV(rad/s)

PDOF(degrees)

LatV(m/s)

BLoc(cm)

1 Baseline 15 4 290 1.5 264.22 LongV − 2.00 13.00 4 290 1.5 264.23 LongV − 1.00 14.004 LongV − 0.50 14.505 LongV − 0.25 14.756 LongV + 0.25 15.257 LongV + 0.50 15.508 LongV + 1.00 16.009 LongV + 2.00 17.00

10 RotV − 1.00 15 3.00 290 1.5 264.211 RotV − 0.50 3.5012 RotV − 0.25 3.7513 RotV − 0.10 3.9014 RotV + 0.10 4.1015 RotV + 0.25 4.2516 RotV + 0.50 4.5017 RotV + 1.00 5.0018 PDOF − 20.0 15 4 270.0 1.5 264.219 PDOF − 15.0 275.020 PDOF − 10.0 280.021 PDOF − 5.00 285.022 PDOF − 2.50 287.523 PDOF + 2.50 292.524 PDOF + 5.00 295.025 PDOF + 10.00 300.026 PDOF + 15.00 305.027 PDOF + 20.00 310.028 LongV − 1.50 15 4 290 0.00 264.229 LongV − 0.75 0.7530 LongV + 0.75 2.2531 LongV + 1.50 3.0032 BLoc − 25.0 15 4 290 1.5 239.133 BLoc − 12.5 251.7

Side impact sled 1.835 0.998Block impact 1.883 0.993Average 1.859 –

herefore, an average ratio of 1.859 was recommended to converthe THUMS pelvic force to impactor force.

ppendix B: Variation study test matrix.

Table B1.

34 BLoc + 12.5 276.735 BLoc + 25.0 289.2

8 nalys

R

B

B

B

B

B

C

C

D

D

D

D

E

E

G

G

G

G

H

H

I

I

K

A.J. Golman et al. / Accident A

eferences

eeman, S.M., Kemper, A.R., Madigan, M.L., Duma, S.M., 2011. Effects of bracing onhuman kinematics in low-speed frontal sled tests. Ann. Biomed. Eng. 39 (12),2998–3010.

elwadi, A., Siegel, J.H., Singh, A., Smith, J.A., Yang, K.H., King, A.I., 2012. Finite ele-ment aortic injury reconstruction of near side lateral impacts using real worldcrash data. J. Biomech. Eng. 134 (1), 011006.

ose, D., Segui-Gomez, M., Crandall, J.R., 2011. Vulnerability of female driversinvolved in motor vehicle crashes: an analysis of US population at risk. Am.J. Public Health 101 (12), 2368.

ouquet, R., Ramet, M., Bermond, F., Caire, Y., Talantikite, Y., Robin, S., Voiglio, E.,1998. Year Pelvis human response to lateral impact. In: Proceedings of the16th International Technical Conference on the Enhanced Safetfy of Vehicles,Windsor, ON, Canada, pp. 1665–1686.

urstein, A.H., Reilly, D.T., Martens, M., 1976. Aging of bone tissue: mechanical prop-erties. J. Bone Joint Surg. Am. 58 (1), 82–86.

avanaugh, J.M., Walilko, T.J., Malhotra, A., Zhu, Y., King, A.I., 1990. Biomechanicalresponse and injury tolerance of the pelvis in twelve sled side impacts. SAE 16.

hang, C.Y., Rupp, J.D., Reed, M.P., Hughes, R.E., Schneider, L.W., 2009. Predicting theeffects of muscle activation on knee, thigh, and hip injuries in frontal crashesusing a finite-element model with muscle forces from subject testing and mus-culoskeletal modeling. Stapp Car Crash J. 53, 291–328.

anelson, K.A., Bolte, J.H., Stitzel, J.D., 2011a. Assessing astronaut injury potentialfrom suit connectors using a human body finite element model. Aviat. SpaceEnviron. Med. 82 (2), 79–86.

anelson, K.A., Chiles, C., Thompson, A.B., Donadino, K., Weaver, A.A., Stitzel, J.D.,2011b. Correlating the extent of pulmonary contusion to vehicle crash parame-ters in near-side impacts. Ann. Adv. Automot. Med. 55, 217–230.

anelson, K.A., Gayzik, F.S., Yu, M.M., Martin, R.S., Duma, S.M., Stitzel, J.D., 2009. Bilat-eral carotid artery injury response in side impact using a vessel model integratedwith a human body model. Ann. Adv. Automot. Med. 53, 271–279.

anelson, K.A., Hoth, J.J., Miller, P.R., Stitzel, J.D., 2007. A semi-automated approachfor measuring splenic injury using computed tomography. Biomed. Sci. Instrum.43, 13–17.

l-Jawahri, R.E., Laituri, T.R., Ruan, J.S., Rouhana, S.W., Barbat, S.D., 2010. Develop-ment and validation of age-dependent fe human models of a mid-sized malethorax. Stapp Car Crash J. 54, 407–430.

ppinger, R.H., Marcus, J., Morgan, R., 1984. Development of dummy and injury indexfor nhtsa’s thoracic side impact protection research program. SAE TechnicalPaper Series (840885).

ayzik, F.S., Hoth, J.J., Daly, M., Meredith, J.W., Stitzel, J.D., 2007. A finite element-based injury metric for pulmonary contusion: investigation of candidate metricsthrough correlation with computed tomography. Stapp Car Crash J. 51, 189–209.

ayzik, F.S., Moreno, D.P., Vavalle, N.A., Rhyne, A.C., Stitzel, J.D., 2012. Develop-ment and validation of a novel full human body finite element model utilizinga multi-modality medical imaging protocol. In: International CrashworthinessConference, Milan, Italy.

olman, A.J., Danelson, K.A., Bolte, J.H., Stitzel, J.D., 2012. Biomechanical assessmentof astronaut injury potential from suit ring elements using a full human bodyfinite element model. In: International Crashworthiness Conference, Milan, Italy.

olman, A.J., Danelson, K.A., Gaewsky, J.P., Stitzel, J.D., 2013. Development and vali-dation of thoracic side impact injury prediction metrics in a human body model.Comput. Methods Biomech. Biomed. Eng. (in press).

ayashi, S., Yasuki, T., Kitagawa, Y., 2008. Occupant kinematics and estimated effec-tiveness of side airbags in pole side impacts using a human FE model withinternal organs. Stapp Car Crash J. 52, 363–377.

ayes, A.R., Vavalle, N.A., Moreno, D.P., Stitzel, J.D., Gayzik, F.S., 2013. Vali-dation of simulated chestband data in frontal and lateral loading using ahuman body finite element model. Traffic Inj. Prev., http://dx.doi.org/10.1080/15389588.2013.799278 (in press).

wamoto, M., Kisanuki, Y., Wantanabe, I., Furusu, K., Miki, K., Hasegawa, J., 2002.Development of a finite element model of the total human model for safety(THUMS) and application to injury reconstruction. In: International ResearchCouncil on the Biomechanics of Injury (IRCOBI), Munich, Germany.

wamoto, M., Nakahira, Y., Kimpara, H., Sugiyama, T., Min, K., 2012. Development of ahuman body finite element model with multiple muscles and their controller forestimating occupant motions and impact responses in frontal crash situations.

Stapp Car Crash J. 56, 231–268.

emper, A.R., McNally, C., Kennedy, E.A., Manoogian, S.J., Rath, A.L., Ng, T.P., Stitzel,J.D., Smith, E.P., Duma, S.M., Matsuoka, F., 2005. Material properties of humanrib cortical bone from dynamic tension coupon testing. Stapp Car Crash J. 49,199–230.

is and Prevention 64 (2014) 1– 8

Kemper, A.R., McNally, C., Pullins, C.A., Freeman, L.J., Duma, S.M., Rouhana, S.M.,2007. The biomechanics of human ribs: material and structural properties fromdynamic tension and bending tests. Stapp Car Crash J. 51, 235–273.

Kent, R., Lee, S.H., Darvish, K., Wang, S., Poster, C.S., Lange, A.W., Brede, C., Lange,D., Matsuoka, F., 2005. Structural and material changes in the aging thoraxand their role in crash protection for older occupants. Stapp Car Crash J. 49,231–249.

Kent, R., Trowbridge, M., Lopez-Valdes, F.J., Ordoyo, R.H., Segui-Gomez, M., 2009.How many people are injured and killed as a result of aging? Frailty, fragility,and the elderly risk-exposure tradeoff assessed via a risk saturation model. Ann.Adv. Automot. Med. 53, 41–50.

Kuppa, S., 2006. Injury Criteria for Side Impact Dummies. National Highway TrafficAdministration, Washington, DC.

Kuppa, S., Eppinger, R.H., Mckoy, F., Nguyen, T., Pintar, F.A., Yoganandan, N., 2003.Development of side impact thoracic injury criteria and their application tothe modified es-2 dummy with rib extensions (es-2re). Stapp Car Crash J. 47,189–210.

Li, Z., Kindig, M.W., Subit, D., Kent, R.W., 2010. Influence of mesh density, corticalthickness and material properties on human rib fracture prediction. Med. Eng.Phys. 32, 998–1008.

Mccalden, R.W., Mcgeough, J.A., Barker, M.B., Court-Brown, C.M., 1993. Age-relatedchanges in the tensile properties of cortical bone. The relative importance ofchanges in porosity, mineralization, and microstructure. J. Bone Joint Surg. Am.75 (8), 1193–1205.

Morgan, R., Marcus, J.H., Eppinger, R.H., 1986. Side impact – the biofidelity of nhtsasproposed ATD and efficacy of TTI. SAE 14.

Mroz, K., Bostrom, O., Pinkorn, B., Wismans, J., Brolin, K., 2010. Comparison of hybridIII and human body models in evaluating thoracic response for various seatbelt and airbag loading conditions. In: Proceedings of the International IRCOBIConference on the Biomechanics of Injury, Hanover, Germany.

Naumann, R.B., Dellinger, A.M., Zaloshnja, E., Lawrence, B.A., Miller, T.R., 2010.Incidence and total lifetime costs of motor vehicle-related fatal and nonfa-tal injury by road user type, united states, 2005. Traffic Inj. Prev. 11 (4),353–360.

NCAC, 2008. Finite element model of Toyota RAV4, model year 1997, version 1.National Crash Analysis Center.

NCAC, 2012. Finite Element Model of Ford Taurus, Model Year 2001, Version 3.National Crash Analysis Center.

NHTSA, 2011. Thoracic Trauma Index Software, Version 10, 11, 10, 27. Washington,DC.

Shah, C.S., Yang, K.H., Hardy, W., Wang, H.K., King, A.I., 2001. Development of acomputer model to predict aortic rupture due to impact loading. Stapp Car CrashJ. 45, 161–182.

Shigeta, K., Kitagawa, Y., Yasuki, T., 2009. Development of next generation human femodel capable of organ injury prediction. Enhanced Safety of Vehicles. NHTSA,Stuttgart, Germany.

Siegel, J.H., Belwadi, A., Smith, J.A., Shah, C., Yang, K., 2010. Analysis of the mecha-nism of lateral impact aortic isthmus disruption in real-life motor vehicle crashesusing a computer-based finite element numeric model: with simulation of pre-vention strategies. J. Trauma 68 (6), 1375–1395.

Stitzel, J.D., Duma, S.M., Cormier, J.M., Herring, I.P., 2002. A nonlinear finite ele-ment model of the eye with experimental validation for the prediction of globerupture. Stapp Car Crash J. 46, 81–102.

Takhounts, E.G., Ridella, S.A., Hasija, V., Tannous, R.E., Campbell, J.Q., Malone, D.,Danelson, K., Stitzel, J., Rowson, S., Duma, S., 2008. Investigation of traumaticbrain injuries using the next generation of simulated injury monitor (simon)finite element head model. Stapp Car Crash J. 52, 1–31.

Urban, J.E., Whitlow, C.T., Edgerton, C.A., Maldjian, J.A., Powers, A.K., Stitzel, J.D., 2012.Motor vehicle crash-related subdural hematoma from real-world head impactdata. J. Neurotrauma.

Vavalle, N.A., Moreno, D.P., Rhyne, A.C., Stitzel, J.D., Gayzik, F.S., 2012. Lateral impactvalidation of a geometrically accurate full body finite element model for bluntinjury prediction. Ann. Biomed. Eng..

WHO, 2009. Global Status Report on Road Safety.Yang, K.H., Hu, J., White, N.A., King, A.I., Chou, C.C., Prasad, P., 2006. Development

of numerical models for injury biomechanics research: a review of 50 years ofpublications in the stapp car crash conference. Stapp Car Crash J. 50, 429–490.

Yoganandan, N., Pintar, F.A., Stemper, B.D., Gennarelli, T.A., Weigelt, J.A., 2007.

Biomechanics of side impact: Injury criteria, aging occupants, and airbag tech-nology. J. Biomech. 40, 227–243.

Zhang, L., Yang, K.H., Dwarampudi, R., Omori, K., Li, T., Chang, K., Hardy, W.N., Khalil,T.B., King, A.I., 2001. Recent advances in brain injury research: a new humanhead model development and validation. Stapp Car Crash J. 45, 369–394.