Printer Friendly

A semi-automated approach to real world motor vehicle crash reconstruction using a generic simplified vehicle buck model.


Computational finite element (FE) modeling of real world motor vehicle crashes (MVCs) is valuable for analyzing crash-induced injury patterns and mechanisms. Due to unavailability of detailed modern FE vehicle models, a simplified vehicle model (SVM) based on laser scans of fourteen modern vehicle interiors was used. A crash reconstruction algorithm was developed to semi-automatically tune the properties of the SVM to a particular vehicle make and model, and subsequently reconstruct a real world MVC using the tuned SVM. The required algorithm inputs are anthropomorphic test device position data, deceleration crash pulses from a specific New Car Assessment Program (NCAP) crash test, and vehicle interior property ranges. A series of automated geometric transformations and five LSDyna positioning simulations were performed to match the FE Hybrid Ill's (HIII) position within the SVM to reported data. Once positioned, a baseline simulation using the crash test pulse was created. A Latin hypercube sample space (9 variables) of 120 simulations was created to vary occupant safety and restraint properties. Sprague and Geers magnitude and phase error factors were used to identify an optimal set of restraint parameters to reconstruct the HIII kinematic and kinetic responses. Using the tuned SVM, event data recorder pulses from real world crashes, and the Total HUman Model for Safety, LS-Dyna simulations were used to reconstruct the occupant-vehicle interactions. In a sample case, stress, strain, and dynamic loads were evaluated to predict rib, sternum, and vertebral injuries sustained by the occupant in the crash.

CITATION: Jones, D., Gaewsky, J., Weaver, A., and Stitzel, J., "A Semi-Automated Approach to Real World Motor Vehicle Crash Reconstruction Using a Generic Simplified Vehicle Buck Model," SAE Int. J. Trans. Safety 4(2):2016, doi: 10.4271/2016-01-1488.


Motor vehicle crashes (MVCs) are the eighth leading cause of death globally, with over 1.2 million deaths [1]. Frontal MVCs are the most common crash mode in the United States, and account for 60% of fatal and 54% of injurious MVCs [2]. While anthropomorphic testing devices (ATDs) and post mortem human subject (PMHS) testing in laboratory conditions are valuable to understanding injury, the time and monetary cost of each ATD/PMHS test severely limits the number of imposed boundary conditions that can be evaluated. Analysis of real-world MVC events allow researchers to both investigate injury patterns and evaluate occupant restraint systems in the fleet in real-world injurious conditions. However, even after expert analysis of crashes, there is often still uncertainty on the exact mechanism of injury, and the effects of pre-crash occupant position on injury outcomes. Benefits of finite element human body models (HBMs) include the ability to calculate specific injury risks in body regions that are not instrumented within ATDs and the ability to quickly assess entire body impact events in many loading conditions. One commonly used HBM capable of predicting organ injuries for whole body impact simulations is the Total HUman Model for Safety (THUMS) [3].

The Crash Injury Research and Engineering Network (CIREN) database contains real world MVC data and has previously been used for computational MVC reconstructions [4]. Golman et al. analyzed the HBM response to predict injury risks across the entire body in a side impact CIREN case, but the reconstruction protocol used an open source 2001 Ford Taurus National Crash Analysis Center (NCAC) full vehicle finite element model (FEM). These methods were therefore limited due the paucity of open source full vehicle FEMs [5, 6].

Another study reconstructed two real world crashes using a methodology similar to this paper, but used a modified Taurus simplified vehicle model (SVM) for reconstruction purposes [7]. While both previous studies reconstructed crashes of sedan vehicles, neither used a vehicle model that was representative of the fleet as a whole.

One challenge of reconstructing CIREN cases is the uncertainty in the occupant's position and posture at the time of the crash [8]. Although CIREN collects information related to the occupant restraint mechanisms and positioning, this data is collected post-crash and is subject to potential error, specifically in occupant restraint parameters. This is due to the priorities of emergency services being focused on occupant safety which often involves moving seat positions to extract the occupant. The objective of this study was to develop a semi-automated approach to reconstruct real world crashes using a standardized buck, information from a CIREN case report, and frontal New Car Assessment Program (NCAP) test data, and subsequently to compare the computed injury risk to a real world crash.


Case Selection

One frontal MVC was selected from CIREN with crash characteristics similar to government crash tests. The crash did not have a rollover event, nor did it include large occupant compartment intrusion.

Ford Escape Case Details

The CIREN frontal crash reconstructed involved an 86 year old, 85 kg, 175 cm (5'9"), belted male driver in a 2012 Ford Escape (CIREN ID 359544180). The Ford Escape struck the back of a 2013 Ford F150 at 0[degrees] PDOF, with an EDR-recorded longitudinal delta-V of 50 km/h (31 mph) resulting in a maximum crush of 43 cm. The frontal airbag deployed in the non-fatal crash and the occupant was documented to be seated "middle to rear track" with an estimated seat back angle of "slightly reclined". The first and second stages of the airbag deployed at 22.5 and 32.5 msec, respectively. The D-Ring was reported to be in the full down position. The occupant sustained multiple injuries including the following: AIS 3 rib fractures, AIS 3 hemothorax, AIS 2 sternum fractures, AIS 2 lumbar vertebral body fracture, and an AIS 2 cervical spinous process fracture (Table 1) [9].

Case Reconstruction Algorithm

The reconstruction process of the CIREN case followed three distinct phases. Phase I involved establishing a vehicle FEM that was suitable for simulating the vehicle environment of a 2012 Ford Escape. This was accomplished by tuning the occupant restraint system of a generic SVM using NHTSA NCAP crash test data. Phase II was completed to automatically position THUMS v4.01 within the previously tuned SVM in a range of pre-crash positions. The seat track, seat back angle, steering wheel angle and telescoping distance, and the D-ring height were the independent variables. Phase III applied the longitudinal EDR pulse from the CIREN crash to each of the previously established potential occupant pre-crash positions. Following these simulations, injury risks for each pre-crash position were assessed using an updated version of the Injury Prediction Post Processor (IPPP), a custom in-house MATLAB GUI (Mathworks Software) [10]. The finite element solver used for each phase was LS_DYNA (MPP, Version 971, R6.1.1., LSTC, Livermore, CA) run on a computer cluster.

Phase I - Simplified Vehicle Model (SVM) Development and Tuning

A FEM of a SVM was adapted, augmented, and tuned to accurately simulate a frontal NCAP crash test of a 2012 Ford Escape. A generic vehicle interior representative of a wide range of modern vehicles was implemented using an adapted version of the vehicle interior developed by Iraeus et al [11]. The generic vehicle geometry developed by Iraeus et al (Umea University) included a seat cushion, seat pan, dashboard, steering wheel and column, door, A-Pillar, BPillar, and center console which represented an average geometry of laser scan data from 14 modern vehicles in the A2MAC1 database [12]. The measurements of the steering column relative to other interior structures in the Iraeus model were used to implement a previously developed steering column, steering wheel, and airbag assembly as well as a seat back. [6, 7, 13]. The steering column was capable of compressing under user defined axial loads. The steering wheel frontal airbag (*AIRBAG_SIMPLE_AIRBAG_MODEL card in LS-DYNA) was based upon an open source NCAC inflating airbag model [14].

NCAP Test Identification and Data Extraction

The 'sisters and clones' vehicle model list was consulted to arrive at a range of vehicle make, models, and years with matching vehicle properties that were subsequently used to query the NHTSA crash test database [15, 16]. A 2011 Ford Escape XLT 4WD SUV (NHTSA Crash Test 7120) was determined to be the most representative of the CIREN case vehicle, with a delta-V of 56.5 km/h (35.1 mph) [17]. Following the identification of the matching crash test, the following data were extracted: Hybrid III (HIII) ATD positioning, eight crash test signals, and lateral side view test video. The signals extracted were belt anchor lap force, belt upper shoulder force, left femur force, right femur force, T6 resultant acceleration, chest deflection, head resultant linear acceleration, and pelvis resultant acceleration using SignalBrowser [18]. The signals were filtered per SAE J211 [19]. The crash test pulse was extracted from the crash test video from camera #7 using Tracker Video Analysis Software (Open Source Physics, Davidson, NC). The longitudinal pulse is shown in Figure 1. This method preserved the vehicle longitudinal, vertical, and pitching kinematics.

FE HIII Positioning

The 50th percentile male HIII ATD FEM (Humanetics, Plymouth, MI) was positioned within the SVM according to the steering column position, seat back angle, pelvis and tibia angles, and nose to rim, chest to steering hub, knee to dash, knee to knee, and ankle to ankle measurements reported in the NCAP report using an algorithmic approach (Figure 2) [20, 21]. The HIII initial positioning was completed by rotating the ATD about the H-point, simulating moving the knees apart, simulating adjustments of the tibia angles, and then simulating the movement of the ankles to the correct position. Next, the SVM steering column angle was adjusted to match the angle specified in the NCAP crash test data by rotating the column about a node at the most forward portion of the part.

The mid-plane of the steering column and the sagittal plane of the HIII model were then aligned. The longitudinal (X) position of the HIII was adjusted to position the lower limbs. The longitudinal (X) position of the HIII was adjusted using an iterative process, due to the complex curvature of the knee bolster surface. During each iteration, a surface to surface measurement was taken between a temporarily created shell element in the knee joint and the right half of the knee bolster. Using this method, the Right Knee to Dash (RKD) distance and Right Knee to Dash Angle (RKDA) were optimized. Once this procedure was completed, a second iterative process was used to match the Left Knee to Dash (LKD) distance. During each iteration, the "separation" measurement was used to determine the shortest distance between the left knee and knee bolster. Based on this distance, the dashboard was rotated about a node on the right side of the knee bolster. This process was terminated when the measured and target distance was less than 0.1 mm.

The relative positions of the HIII upper body were determined based on the steering wheel diameter and reported nose to rim distance, nose to rim angle, chest to steering hub distance, and steering wheel angle. The necessary relative horizontal distance and angle between the chest and nose were calculated and the neck joint was rotated about the local Y-axis in the LS-PrePost Dummy Positioner until the relative horizontal distance was matched. After the upper body segments of the HIII model were repositioned, the steering wheel and column positions were refined by translating the entire steering column and wheel assembly.

The HIII was then translated upwards until there was no penetration with the seat and allowed to settle into the seat by locking the joint movement and applying gravity. At each d3plot state of the settling simulation, the angle of the head with respect to the horizon was measured. Once this initial angle deviated by more than 1.5[degrees], the model was determined to be in the correct position and node locations were saved. This corresponded with the H-point location that minimized the distance to the floor during this phase (Figure 3).

Following this step, the seat back was rotated from the fully reclined position upwards 75[degrees] over a period of 300 ms. A node on the occupant's upper back and a node of the upper seat back cushion were tracked, and the state at which this distance was less than 10 mm was used for the final node locations. This ensured that the seat was in a position similar to a crash test using a published positioning procedure [22]. The occupant was then automatically belted within the SVM. The positioned HIII model in the pre-crash state is shown in Figure 4.

Latin Hypercube Design of Experiments

The occupant restraint parameters were subsequently used in a variation study using a Latin Hypercube Design (LHD) of experiments. The LHD is an effective space filling design used in experimental parameter studies [23]. In a LHD, each parameter has as many levels as there are experiments in the design. The levels are spaced evenly from the lower bound to the upper bound of the parameter. The Latin Hypercube optimization method used in this study was the optimumLHD function in R (The R Foundation). The previously described SVM and HIII set up was used in 200 different simulations with 9 different independent variables. The model's kinematics were driven by the video tracker data. The nine independent occupant restraint variables were divided into five systems: frontal airbag, seatbelt, knee bolster, steering column, and seat cushion. For the belt, maximum pretensioner force, belt buckle dynamic friction coefficient, and load limiting force were varied. For the frontal airbag, both the mass flow rate and vent area were included as independent variables. The mass flow rate was a function of time with a right skewed shape (Figure 5). A scale factor on the ordinate of this mass flow rate was used as an independent variable. The knee bolster had two independent variables: right and left knee bolster surface thickness. The seat cushion modulus was varied to account for potential different seat stiffnesses from various assemblies. Additionally, the shear bolt fracture force in the steering column was also varied.

Following the completion of the 200 simulations, signals were extracted from locations in the FE HIII representing the eight extracted from the NHTSA crash test. Sprague and Geers error analysis was used to compare the paired signals [24]. This analysis calculates both the magnitude (M) and phase (P) error for each signal. The comprehensive (C) error was also calculated for each signal according to equations 1, 2, 3, 4, 5.






Where m(t) corresponds to the metric data for the simulation model and e(t) corresponds to the metric data from the experimental test. The M and P error factors were calculated over a time period from [t.sub.1] [less than or equal to]t[less than or equal to][t.sub.2] For this study, the time range was from 0 to 100 ms. Data after this time period described the ATD motion during the rebound phase.

Based on these definitions, M, P, and C were standardized measures of the difference between the signals of the experimental tests and simulation models. Smaller values represent more similar signals. P can range from 0 to 1, while M can theoretically range from -1 to [infinity]. However, typical values range was from -1 to 1. The IPPP was used to batch this operation for all 200 simulations. In order to arrive at an overall metric of similarity of the SVM to NCAP data, a total body comprehensive error factor was developed: [C.sub.TotalBody] (Eq 6). This metric was developed by averaging the right and left femur errors and the seat belt errors and then weighting the results equally to the head, T6, and pelvis [C.sub.metric] values.


The top three simulations were then further analyzed with each parameter across the three simulations averaged to arrive at a hypothesized "optimal" tuned SVM. A final simulation with the tuned SVM was subsequently performed to test this hypothesis.

Phase II - THUMS Scaling and Variation Study

After tuning the SVM to have a similar frontal crash response as a 2012 Ford Escape, the next step in the crash reconstruction was to implement THUMS into the SVM and run a positioning variation study. Because the pre-crash position is typically uncertain in real world crashes, a set of 120 pre-crash occupant positions and adjustable vehicle parameters were created.

THUMS Occupant Scaling

Since the occupant in the CIREN case and THUMS v4.01 are neither the same weight nor height, the model was scaled. One option was to scale THUMS to the correct height by equally scaling the lengths of the HBM. However, this would ultimately lead to incorrect mass (12.6% error). Similar efforts to scale the mass would yield a large height error. Thus, a set of equations were derived to calculate a length-scaling factor that minimized both the length and mass errors without compromising the model geometry and properties (Eq 7-9).

height factor = case occupant heiqht/THUMS model height (7)


scale factor = heiqht factor + mass factor/2 (9)

Solving these equations using the case occupant's mass (84 kg) and height (175 cm) as well as THUMS height (178 cm) and mass (73 kg), yielded a length scale factor of 1.016. This scale factor was used to increase THUMS length in the X-, Y-, and Z- directions simultaneously. This yielded a scaled THUMS model measuring 181.7 cm (3.81% error) and 76.6 kg, (-4.29% error).

Initial THUMS Positioning Simulations

Once the THUMS occupant model was scaled, two simulations were performed to place THUMS into the tuned SVM. A 0.5 G load was applied simultaneously to THUMS in the X and Z directions. The nodes were exported from the model at the last d3plot state before the head had rotated more than 2[degrees] with respect to original angle. 2[degrees] was chosen as the indicator for THUMS since the joints were more flexible than the ATD and rotation presented itself earlier. Next, the hands and feet were pulled to the steering wheel and foot pedals, respectively by temporarily turning the corresponding bones to rigid materials and using the *BOUNDARY_PRESCRIBED_MOTION card in LS-DYNA while gravity was still turned on. The initial position of THUMS within the SVM as well as the baseline position reported in the CIREN case are shown in Figure 6.

THUMS Occupant Positioning Variables

Five variables were defined to vary the THUMS' occupant position across simulations for each case reconstruction: 1) steering column angle, 2) steering column fore/aft position, 3) seat back angle, 4) seat track position, and 5) D-Ring anchor height. The minimum and maximum steering column angle, as well as minimum and maximum steering column fore/aft position, and seat track positions were gathered from the NCAP crash test report. These parameter ranges were sampled using an LHD of 120 simulations. A simulation was performed to modify the seat back angle and seat track position for each occupant position. After THUMS was repositioned, an automatic belting code was employed for each repositioned occupant.

Phase III - Applied EDR Pulse and Post-Processing

Following the positioning of the THUMS model into the 120 different positions, the longitudinal EDR crash pulse (Figure 1) was used to drive each simulation.

Each simulation was carried to completion at 200 msec. After simulations were inspected for quality, the IPPP was used to extract injury metrics and calculate injury risks. Chest deflection (also defined as chest compression), lumbar vertebral bending moment, and cross sectional loads were evaluated for each simulation. The simulated injury risks were then compared to the injuries experienced by the CIREN occupant. Chest deflection was evaluated to estimate an injury risk probability of rib and sternum fractures and hemothorax that were present in the CIREN case occupant. To compare the reconstructed occupant's lumbar response to the CIREN occupant, local forces and moments were measured in the L2, L3, and L4 vertebrae (Figure 7).

In addition to evaluating the injury risks for the occupant position listed in the CIREN report, the risk for each injury was compared across pre-crash positions.


Simplified Vehicle Model Tuning

The FE HIII was seated in the SVM prior to tuning in a position similar to the NCAP report. Since the SVM was based on average fleet vehicle geometry, not all measurements could be matched simultaneously. A list of positioning variables and ATD positioning before SVM tuning can be found in Table 2. The measurements that were closest were ankle to ankle distance, tibia angle, chest to steering hub, and steering wheel angle. Pelvic angle and knee to knee distance were less accurately matched. Ultimately, the initial position of the FE HIII within the geometry of the SVM was close to the HIII in the NCAP test. The average error for distance measurements was 1.06 cm. For angles, the average error was 4.43[degrees].

Following the initial positioning of the HIII into the SVM, the next step was to simulate the NCAP crash test kinematics while varying the occupant restraint parameters. Each simulation was ranked using the [C.sub.TotalBody] metric. The top three simulations ([C.sub.TotalBody] = 0.573, 0.584, 0.590) were further analyzed by plotting [C.sub.TotalBody] and [C.sub.metrtc] for each signal, against each parameter. An example of these plots is shown for the total body comprehensive error factor versus the left knee bolster thickness scale factor (Figure 8). It is evident that for this example, there exists a range of reasonable values that can dramatically improve the ability of the SVM to replicate the response of the real world vehicle.

Following the analysis of the top three simulations, a hypothesized optimal simulation was generated by averaging the parameters used in the top three simulations previously mentioned. The resulting occupant restraint parameters, as well as the range of tested parameters are listed in Table 3. A comparison between the simulated and NCAP test signals using the hypothesized optimal parameters can be seen in Appendix A.

The ATD responses in the NCAP test compared favorably to the simulation responses. Cursory analysis shows that each signal is on the same order of magnitude. The magnitude (M), phase (P), and comprehensive (C) Sprague and Geers error for each signal is listed in Table 4. Lower values represent closer matches, with 0 being an exact match. The magnitude error is below 0.25 for each signal and the phase error is below 0.45 for each signal. The majority of the discrepancies were in the femur force signals, though the forces were on the lower end of the injury risk curve spectrum. The comprehensive error factors are highest in the knee-thigh-hip region as well. Overall, however, this tuned SVM performed well, and the optimized parameters resulted in a lower total body comprehensive error factor than any of the simulations generated by the LHD alone with [C.sub.TotalBody] = 0.553.

Effects of Pre-Crash Positioning and Injury Risk Analysis

Chest deflection was used as a measure for assessing potential rib and sternum fractures and other chest injuries. More rearward seated simulated occupants tended to have a greater chest deflection (compression) than forward seated occupants (Figure 9). Analysis of sternum trabecular and cortical bone max principal stress show values exceeding ultimate stresses for older occupants found in literature (Figure 10). These thresholds are 117 MPa and 2 MPa for cortical and trabecular bone, respectively [25].

The pre-crash occupant position had a notable effect on the peak loads experienced in the lumbar vertebrae during the crash event. The relationships between the lumbar loads and the seat track position and seat back angle were plotted in Figure 11 and Figure 12 across all completed simulations. Time history data for the baseline case was plotted in Figure 13. The maximum compressive force in L3 ranged from 537 to 1874 N across the simulated positions. As the seat track was moved forward, the occupant sustained greater compressive forces in the lumbar vertebrae. This trend with seat track position was the opposite of that seen in the head and chest regions. The most reclined occupant positions tended to have higher compressive forces in the lumbar vertebrae than the moderately reclined occupant positions. The most upright occupants also had increased compressive lumbar forces. The more reclined occupants tended to have higher peak bending moments of the lumbar spine during the crash event.


While the NHTSA crash test did not exactly match the CIREN crash, the EDR delta-V was very similar to that of the NCAP case. The initial position of the FE HIII ATD was subject to some error; the primary influence was the SVM geometry. To match the ATD position measurements exactly, the geometry of the SVM would need to mirror the specific vehicle's geometry exactly. The pelvic angle error was due to the rotation of the HIII during settling simulations. The knee to knee distance was less accurate than the ankle to ankle distance because the ankle to ankle movement was simulated after the knee movement simulation.

Many of the ATD signals were matched closely. However, after analysis of the comprehensive errors, the need for increased variability of the knee bolster was evident. This was the primary contributor of error in the femur loads and in the future could be alleviated by introducing a material model elastic modulus scale factor in conjunction with the thickness scale factor or a new type of contact definition. Pelvis acceleration matched the dual-peaks in the NCAP environment, but exceeded the maximum value by approximately 30 G's. This may be due to the increased femur loads propagating force to the pelvis. Chest deflection in the simulated environment was higher than in the NCAP environment, but resulted in less than 1 cm of error. Both belt forces were in good agreement with the NCAP data signals and although the head acceleration was overestimated, it was below injurious thresholds. Expanding the LHD in both the number of variables and simulations may increase the fidelity of the occupant response in the SVM.

By developing crash reconstruction and occupant repositioning algorithms for a generic simplified vehicle interior model, injury patterns as a function of pre-crash occupant position were evaluated. Injuries to the thoracic region, including rib and sternum fractures, were correlated to increased chest displacements. More rearward occupant positions had increased chest deflections, indicating an increased likelihood of chest injury. This may be due to increased interaction with the shoulder belt before the distributed load of the airbag decelerated the occupant. While the chest deflections were not noticeably increased for reclined occupants, greater stresses than reported failure thresholds in the sternum were identified.

The injury prediction trends varied for the lumbar region. The most rearward seated occupants tended to have lower compressive forces in the L2, L3, and L4 vertebrae. The most reclined and most upright occupants tended to have increased compressive loading compared to the moderately reclined occupant positions. Meanwhile the greatest lumbar vertebrae bending moments occurred in the most reclined occupants. Similar injury metric evaluations in future studies may help elucidate concurrent injury mechanisms across body regions in specific occupant scenarios.


The SVM was tuned using a NCAP crash test with a delta-V of 56.5 km/h in order to recreate a real world crash with a delta-V of 50 km/h. As such, large deviations from this delta-V range would likely lead to a need for re-tuning. Under-ride was not coded into the CIREN database, but the crush above the bumper was greater than at bumper level (Appendix B). Since there was not occupant compartment intrusion, it was deemed acceptable to drive the tuned SVM by the kinematics extracted from the EDR.

While human body models have been used in the past to predict injuries in simulated scenarios, the finite element method has some inherent limitations. First, human models lack the detailed musculature and muscle activation patterns of living humans. This may affect the kinematics of the occupant before and after impact onset as well as the potential muscle tissue injuries experienced by real world MVC occupants. Furthermore, the model may not capture the intricacies involved in pre-impact bracing.

As it is, the model has the ability to match polarity of force in the femurs, but may become too stiff during deformation which leads to decreased interaction time between the knees and bolster. This may be improved by implementing a new knee bolster interaction calculation. Another notable departure from the Escape to the SVM was the design of the seat. Both the NHTSA crash test and CIREN crash vehicles featured a motorized seat bottom while the SVM included a much simpler manually adjusted tray. This may have limited the ability to perfectly match the pelvis accelerations.


In summary, a robust, semi-automated crash test reconstruction methodology was developed. This method used minimal input by the user to automatically position the ATD according to the NCAP crash test data, tune the SVM to that of the target vehicle, and carry out the reconstruction of the THUMS HBM in the tuned SVM. The tuning procedure yielded an FE SVM that was representative of the target vehicle by comparing eight signals using Sprague and Geers error analysis. A baseline simulation was completed using the tuned SVM. This semi-automated method can be used in the future to more-efficiently reconstruct frontal MVCs. By doing so, the knowledge of injury mechanisms in MVCs will be extended and new injury metrics can be developed. This tool could also be used by CIREN partners in conjunction with their case-by-case analysis.


[1.] WHO, "Global Status Report on Road Safety 2013," World Health Organization, Luxembourg2013.

[2.] NHTSA, "Traffic Safety Facts 2012: A Compilation of Motor Vehicle Crash Data from the Fatality Analysis Reporting System and the General Estimates System," National Highway Traffic Safety Administration, Washington, DC2014.

[3.] Shigeta K., Kitagawa Y., and Yasuki T., "Development of next generation human FE model capable of organ injury prediction," in Enhanced Safety of Vehicle Conference, Stuttgart, Germany, June, 2009. pp. 15-18.

[4.] Schneider L. W., Rupp J. D., Scarboro M., Pintar F., Arbogast K. B., Rudd R. W., Sochor M. R., Stitzel J., Sherwood C, Macwilliams J. B., Halloway D., Ridella S., and Eppinger R., "BioTab--a new method for analyzing and documenting injury causation in motor-vehicle crashes," Traffic Inj Prev, vol. 12, pp. 256-65, Jun 2011.

[5.] Golman A. J., Danelson K. A., and Stitzel J. D., "Robust human body model injury prediction in simulated side impact crashes," Comput Methods Biomech Biomed Engin, pp. 1-16, Jul 9 2015.

[6.] Marzougui D., Samaha R. R., Cui C., and Kan C. D., "Extended Validation of the Finite Element Model for the 2001 Ford Taurus Passenger Sedan," National Crash Analysis Center, George Washington University, Ashburn, VA2012.

[7.] Gaewsky J. P., Weaver A. A., Koya B., and Stitzel J. D., "Driver Injury Risk Variability in Finite Element Reconstructions of Crash Injury Research and Engineering Network (CIREN) Frontal Motor Vehicle Crashes," Traffic Inj Prev, vol. 16 Suppl 2, pp. S124-31, Oct 8 2015.

[8.] DanelsonK. A. and Stitzel J. D., "Finite element model prediction of pulmonary contusion in vehicle-to-vehicle simulations of real-world crashes," Traffic Inj Prev, vol. 16, pp. 627-36, 2015.

[9.] "Abbreviated Injury Scale 2005," 2005.

[10.] Golman A. J., Danelson K. A., Miller L. E., and Stitzel J. D., "Injury prediction in a side impact crash using human body model simulation," Accid Anal Prev, vol. 64, pp. 1-8, Mar 2014.

[11.] Iraeus J., "Stochastic Finite Element Simulations of Real Life Frontal Crashes," Doctoral Thesis, Department of Surgical and Perioperative Sciences, Umea University, Umea, 2015.

[12.] A2MAC1 Automotive benchmarking [Online]. Available:

[13.] Gaewsky J. P., Danelson K. A., Weaver C. M., and Stitzel J. D., "Optimization of a simplified automobile finite element model using time varying injury metrics," Biomed Sei Instrum, vol. 50, pp. 83-91, 2014.

[14.] Bedewi N. E., Marzougui D., and Motevalli V., "Evaluation of parameters affecting simulation of airbag deployment and interaction with occupants," Int J Crashworthiness, vol. 1, pp. 339-354, 1996.

[15.] AndersonG. C., "Vehicle Year & Model Interchange List (Sisters & Clones List)," 2011.

[16.] NHTSA. NHTSA Vehicle Crash Test Database [Online]. Available:

[17.] NHTSA, "NCAP-MGA-2011-034," MGAResearch Corporation, Burlington, WI2010.

[18.] N. N. R&D, "SignalBrowser," in VSR_RDDBsoftware, v5.15.3.31 ed.

[19.] SAE International Surface Vehicle Recommended Practice, "Instrumentation for Impact Test - Part 1 - Electronic Instrumentation," SAE Standard J211-1, Rev. Dec. 2003.

[20.] "H350 Adult Dummy Model LS-Dyna," Humanetics Innovative Solutions, Inc.2012.

[21.] FTSS, "LS-DYNA Model of the Hybrid III 50th Percentile Male Dummy User Manual," First Technology Safety Systems, Plymoth, MI2010.

[22.] Manary, M., Reed, M., Flannagan, C., and Schneider, L., "ATD Positioning Based on Driver Posture and Position," SAE Technical Paper 983163, 1998, doi: 10.4271/983163.

[23.] Stocki R., "A method to improve design reliability using optimal latin hypercube sampling," Computer Assisted Mechanics and Engineering Sciences, vol. 12, p. 393, 2005.

[24.] Sprague M. and Geers T., "A spectral-element method for modelling cavitation in transient fluid-structure interaction," International journal for Numerical Methods in Engineering, vol. 60, pp. 2467-2499, 2004.

[25.] Schoell S. L., Weaver A. A., Urban J. E., Jones D. A., and Stitzel J. D., "Development and Validation of an Older Occupant Finite Element Model of a Mid-Sized Male for Investigation of Age-related Injury Risk," Stapp Car Crash Journal, vol. 59, 2015.


Address correspondence to

Joel D. Stitzel

Virginia Tech-Wake Forest University Center for Injury


Wake Forest University School of Medicine

575 N Patterson Ave., Winston-Salem, NC 27101, USA


Views expressed are those of the authors and do not represent the views of any sponsors. Funding for this project was provided by Toyota's Collaborative Safety Research Center. The authors would like to thank Johan Iraeus and Mats Lindkvist for providing the SVM geometry used in the case reconstruction. Computations were performed on the Wake Forest University DEAC Cluster, a centrally managed resource with support provided in part by the University, and the Blacklight system at the Pittsburgh Supercomputing Center (PSC). The authors acknowledge Ian Marcus and Bharath Koya for implementing the seat and steering column into the SVM and Logan Miller, Nick White, and Mireille Kelley for the implementation of injury metrics in THUMS. The authors acknowledge Ryan Barnard for his programming contributions. Views expressed are those of the authors and do not represent the views of any of the sponsors.


ATD - anthropomorphic test device

CIREN - Crash Injury Research and Engineering Network

EDR - event data recorder

FE - finite element

FEM - finite element model

HIC - Head Injury Criterion

HIII - Hybrid III

IPPP - Injury Prediction Post Processor

LKD - left knee to dash

MVC - motor vehicle crash

NCAC - National Crash Analysis Center

NCAP - New Car Assessment Program

NHTSA - National Highway Traffic Safety Administration

PDOF - principal direction of force

PMHS - post mortem human subject

RKD - right knee to dash

RKDA - right knee to dash angle

SVM - simplified vehicle model

THUMS - Total HUman Model for Safety




Crush Profile       Unit  C1    C2    C3    C4    C5    C6    Delta-V

NHTSA Test #7120    cm    50.9  52.8  54.4  56.8  54.3  53.1  56.5
CIREN Above Bumper  cm    35    36    40    38    39    39    km/h
CIREN Bumper        cm     7    28    39    26    10     0    50 km/h
CIREN Average       cm    21    28    39    26    25    20

Derek Jones, James Gaewsky, Ashley Weaver, and Joel Stitzel Wake Forest Univ.

Table 1. Injuries sustained in 2012 Escape MVC. AIS Codes paired with
description [9].

AIS Code  Injury Description

450203.3  Rib fractures (left 3-4 and right 5-7, 10)
442200.3  Right hemothorax
450804.2  Central sternum fracture
450804.2  Manubrium fracture
650632.2  L3 compression fracture
650218.2  C4 spinous process fracture

Table 2. ATD positioning values for both NCAP test and FE HIII in SVM.

                            NCAP Crash Test         Simulated ATD
                                                    Position in SVM

Steering Wheel Angle   SWA   66.1[degrees]          64.7[degrees]
Steering Column Angle  SCA   23.9[degrees]          25.3[degrees]
Nose to Rim            NR   384mm @ 12.9[degrees]  378 mm @ 2.4
Chest to Steering Hub  CS   305 mm                 301 mm
Left Knee to Dash      KDL  136 mm                 154 mm @ 24.3
Right Knee to Dash     KDR  127 mm                 144 mm @ 30.5
Pelvic Angle           PA    24.9[degrees]          21.2[degrees]
Tibia Angle            TA    46.9[degrees]          45.1[degrees]
Knee to Knee           KK   358 mm                 340.2 mm
Ankle to Ankle         AA   329 mm                 329.8 mm

Table 3. Tuning parameters for the SVM. The "selected" column
represents the values that tuned the SVM to a 2012 Escape and
subsequently were used for Phase II and III.

Parameter                          Min       Selected  Max

Maximum Pretensioner Force (N)     3250      3909       4250
Load Limiting Force (N)            2950      3521       3800
Peak Frontal Airbag                   1.7       2.4        3.3
Inflation Rate (kg/s)
Frontal Airbag Vent Area (mm)       900      2014       2200
Belt Buckle Friction Coefficient      0.075     0.105      0.150
Seat Cushion Modulus Scale Factor  4000      7427      16000
Thickness of Right Side of            0.5       1.402      1.5
Knee Bolster Surface (mm)
Thickness of Left Side Knee           0.3       0.675      1.8
Bolster Surface (mm)
Shear Bolt Fracture Force (N)      3800      4981       7500

Table 4. Sprague and Geers error factors for the tuned SVM.

Signal                              M       P      C

Belt Anchor Lap Force (CFC 60)      -0.115  0.059  0.129
Belt Upper Shoulder Force (CFC 60)   0.191  0.069  0.203
Left Femur Force (CFC 600) Z         0.026  0.339  0.340
Right Femur Force (CFC 600) Z        0.111  0.445  0.459
T6 Accel (CFC 180) Res              -0.132  0.052  0.142
Chest Deflection                    -0.244  0.060  0.220
Head Accel (CFC 1000) Res           -0.224  0.071  0.235
Pelvis Accel (CFC 1000) Res         -0.028  0.205  0.207
COPYRIGHT 2016 SAE International
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2016 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Jones, Derek; Gaewsky, James; Weaver, Ashley; Stitzel, Joel
Publication:SAE International Journal of Transportation Safety
Article Type:Report
Date:Jul 1, 2016
Previous Article:The accuracy of an optimized, practical close-range photogrammetry method for vehicular modeling.
Next Article:Influence of pre-impact pedestrian posture on lower extremity kinematics in vehicle collisions.

Terms of use | Privacy policy | Copyright © 2019 Farlex, Inc. | Feedback | For webmasters