AN INVESTIGATION OF BRAIN INJURY RISK IN VEHICLE CRASHES (SECOND REPORT)

Transcription

1 AN INVESTIGATION OF BRAIN INJURY RISK IN VEHICLE CRASHES (SECOND REPORT) Takahiro, Kikuchi Kaoru, Tatsu Chinmoy, Pal Shigeru, Hirayama Nissan Motor Co., Ltd. Japan Paper Number ABSTRACT In 2013, an angular velocity based brain injury criterion BrIC, has been proposed by the National Highway Traffic Safety Administration (NHTSA) for consumer vehicle safety assessment tests. In this study, the effect of duration of angular velocities on the predictor s precision was examined. The cumulative strain damage measure (CSDM) and the maximum principal strain were calculated with the data of 445 anthropomorphic test device (ATD) in various vehicle crash tests conducted by NHTSA and the Insurance Institute for Highway Safety (IIHS) using the Simulated Injury Monitor (SIMon ver. 4.0), a finite element model of human brain developed by NHTSA s research institute. The test dataset which composed of different risk levels of brain injury CSDM, MPS, BrIC and their corresponding angular velocities and durations were classified using Self-Organizing Maps (SOMs) combined with hierarchical clustering. The result showed that the differences of the probability of the risks between CSDM, MPS and the corresponding BrICs might be larger when the peak values of angular velocities were higher and the corresponding time durations were shorter. INTRODUCTION The level of head injury risk of occupants in vehicle crashes is usually evaluated with HIC which is calculated using three components of linear head acclerations of ATD. Therefore, it is not possible to predict the brain injury risk caused by head rotational motions by HIC. Takhounts et al. proposed a kinematically based brain injury criterion, BrIC, to be used in regulatory or consumer safety vehicle safety assessment tests [1]. It is calculated with the peak values of angular velocities around three axis. If the time durations of critical angular velocities around three axis could be adujsted for loading signals to head, the coefficient of determination between CSDM and BrIC was not improved from the original formulation [2]. In our previous study, multi-variable regression analysis confirmed that, in addtion to the peak values of angular velocities, incorporating the peak values of angular acceleration around each axis would improve the accuracy of the predictor [3]. The purpose of this study was to examine the effect of head angular velocities around each axis and the corresponding time durations on the accuracy of BrIC. Data were obtained on 445 ATDs in vehicle crash tests conducted at NHTSA and IIHS. The probability of AIS 4+ brain injury risks based on CSDM, MPS, BrIC and their corresponding peak values of angular velocities and their time durations were classified and analyzed visually with SOMs, a kind of neural network algorithm, combined with hierarchical clustering alogorithm. METHODS Data set and variables Frontal and lateral vehicle crash test data for 445 ATDs used in this study are shown in table 1. Table 1. Test conditions and number of ATD Crash test condition No. of ATDs Frontal Frontal RB 84 Offset DB 20 Small overlap RB 132 Oblique offset MDB 57 Lateral FMVSS 214 MDB 64 IIHS MDB 46 Pole 38 Vehicle to vehicle 4 RB : Rigid barrier, DB : Deformable barrier MDB: Moving deformable barrier Kikuchi 1

2 Angular Velocity [rad/s] These were obtained from NHTSA s [4] and IIHS s site [5]. The SIMon code developed by NHTSA was used to calculate CSDM and MPS with these test data. Strain threshold of 0.25 was used to calculate the CSDM for each test [1]. Probabilities of AIS 4+ brain injury risks were then calculated with these two metric and two probabilities of AIS 4+ brain injury were calculated by CSDM and MPS based BrIC with the formulation in the literature [1]. In addition, the peak values of angular velocities around three axis of dummy head and the corresponding time durations were calculated for each test to classify dummy data. A time duration of angular velocity used in this study was defined as shown in figure 1 [2]. Tleft and Tright shown in Figure 1 are the closest intersection of time axis to the maximum value of angular velocity. The time duration is Tright Tleft Maximum of angular velocity T left T right Time [msec] Time duration (DT) Figure 1. Duration of angular velocity 12 variables used for classification of dummy data are shown in table 2. These variables were nondimensionalized by dividing them with the range from minimum to maximum after subtracting the minimum value for the corresponding variables when Euclidian distances as proximity of dummy data were calculated. Table 2. Variables for classification of ATD data No Variable Description Name 1 CSDM Cumulative strain damage measure 2 MPS Maximum principal strain 3 BrIC Brain rotational injury criteria 4 DTx Duration of angular velocity around fore-aft axis 5 DTy Duration of angular velocity around horizontal axis 6 DTz Duration of angular velocity around vertical axis 7 PCSDM The probability of AIS 4+ brain injury based on CSDM 8 PMPS The probability of AIS 4+ brain injury based on MPS 9 PBrIC_CSDM The probability of AIS 4+ brain injury predicted by CSDM based BrIC 10 PBrIC_MPS The probability of AIS 4+ brain injury predicted by MPS based BrIC 11 DIFF_CSDM Difference between PCSDM and PBrIC_CSDM 12 DIFF_MPS Difference between PMPS and PBrIC_MPS Visualizing test data with SOMs [6] A schematic diagram of SOM is shown in Figure 2. SOMs were used to visualize in which tests probabilities of AIS 4+ brain injury based on CSDM and MPS were well-predicted by BrIC and also identified the tests where they were not wellpredicted. The ATDs data were non-linearly mapped on a two-dimensional layer where the locations of the input data were determined based on the weighted Euclidian distances. The weighted values of variables from #7 to #12 in table 2 were set to zero to prevent highly correlated variables from affecting the SOM results. ATD data in each test Euclidian distances Output layer RESULTS d 1 d 2 Unit Figure 2. Self-Organizing Maps Comparison of level of brain injury risk Figure 3(a) shows the comparison of brain injury risk predicted by CSDM (vertical axis) and that of BrIC (horizontal axis), while Figure 3(b) shows the comparison of brain injury risk predicted by MPS (vertical axis) and that of BrIC (horizontal axis). d i d n Kikuchi 2

3 Porobability of AIS 4+ based on MPS Porobability of AIS 4+ based on CSDM The results inside the dotted ellipse indicates that PCSDM and PMPS values were higher than those of PBrIC. Therefore, BrIC underestimated the levels of brain injury risks compared with CSDM and MPS in these tests. The accuracy of the prediction in such severe loadings will be important when there is a possibility of high brain injury risk in a vehicle safety performance test. Therefore, the effect of the peak level of head angular velocities around each axis and their corresponding time durations on the accuracy of BrIC were thoroughly examined Probability AIS4 + predicted by CSDM based BrIC (a) Probability AIS4 + predicted by MPS based BrIC 6 clusters were found to be appropriate to analyze the effect of variables on the precision of BrIC. Table 3 shows the number of dummy in each cluster. Table 3. Number of ATD in each cluster No. of No. of Cluster dummy data Figure 5 shows the output layers for each variable such as PCSDM, PMPS, etc. Black dots in each map represent ATDs data in all tests and are located in the same positions in all maps. The values of the variables in each region increase as the color of the regions becomes warmer. A B (b) Figure 3. Comparison results of brain injury risk predicted by (a) CSDM and BrIC; (b) MPS and BrIC Cluster analysis of dummy data The name and cluster locations are shown in Fig. 4. Cluster 1 Cluster 2 Cluster 3 C D Cluster 6 Cluster 5 Cluster 4 Figure 4. Self-Organizing Maps and cluster names Figure 5. Self-Organizing Maps Kikuchi 3

4 Average value (normalized) Relatively higher levels of brain injury risk of the ATDs data based on CSDM, MPS and their corresponding BrICs gathered on the left side of these maps and classified into cluster 1, 2 and 6 (marked as A). The output layers of the variable DIFF_CSDM and DIFF_MPS showed that clusters 1, 2 and 6 had relatively higher values of this variable than the other clusters (marked as B). The output layers of the variable DTX, DTY, DTZ showed that clusters 1, 2 and 3 had data which had relatively shorter time duration of angular velocities than the other clusters (marked as C), while tests which had relatively longer time durations of angular velocities were classified into cluster 6 (marked as D). Here comparing data from cluster 1, 2, 3 would be helpful to clarify the mechanisms why the probability of AIS 4+ based on CSDM and MPS were not predicted well by corresponding BrIC values in some tests like as shown by dotted ellipse in Figure 1, compared to those well-predicted in other tests like cluster 3. In addition, the number of test data that belonged to those clusters were comparatively large except cluster 6 which contained only five test data. Figure 6 shows the average values of twelve variables for cluster 1, 2 and 3. The values of DIFF_CSDM for cluster 1 and 2 were the almost same. They were approximately three times as that of cluster 3. The value of DIFF_MPS of cluster 1, on the other hand, was larger than that of cluster 2 and three times larger than that of cluster 3 (marked as E). The average values of time duration of angular velocities around x and y axes in cluster 1 and 2 were shorter than those of cluster 3 in this manner while that of angular velocity around z were close to each other (marked as F) P CSDM P BrIC_CSDM P MPS DIFF_CSDM P BrIC_MPS DT X DT Y DIFF_MPS CSDM MPS BrIC Frequency Frequency Frequency Cluster 1 Cluster 2 Cluster 3 E DT Z Figure 6. Average values of each cluster Figure 7 shows the distributions of time duration of angular velocities around three axis for cluster 1, F 2 and 3. There were some tests in which the time durations of angular velocities around three axis were extremely short (marked as dotted ellipse) Figure 7. Distribution of duration of angular velocities around each axis DISCUSSION Cluster 1 Cluster 2 Cluster 3 DTx [msec] DTy [msec] DTz [msec] Effect of time duration of angular velocities on the precision of BrIC Figure 5 shows test data which had relatively higher probability of AIS 4+ brain injury based on CSDM, MPS and BrIC were classified to cluster 1 and 2. The number of data in cluster 1 and 2 were 26 and 40 respectively and not so few. Moreover, those clusters had the tests which had relatively higher Kikuchi 4

5 Angular Velocity [rad/s] Probability of brain injury Angular Velocity [rad/s] Probability of brain injury DIFF_CSDM and DIFF_MPS values, indicating that BrIC was less accurate in predicting the brain injury risk based on CSDM and MPS in these clusters (marked as B). In Figure 6, a comparison of the average values of cluster 1, 2 and 3 indicated that the precision of BrIC of cluster 1 and 2 which had higher probability of AIS 4+ brain injury and shorter time duration of angular velocities were worse than that of cluster 3. Based on these findings, the differences of the probability of risks among CSDM, MPS and the corresponding BrICs might be larger when the data have higher peak values of angular velocities and shorter time durations. Such typical examples were compared and shown in Figure 8(a), 8(b), in which the upper graph corresponds to the result of shorter time duration and the lower graph is related to the result of relatively longer time duration. They had close values of probability of AIS 4+ brain injury predicted by CSDM and MPS based BrIC (marked as dotted ellipse in Figure 8(a), 8(b)) (a) (b) -1.0 Figure 8. Typical cases with (a) shorter (b) longer time duration of angular velocities PCSDM PBrIC_CSDM Time [ms] Time [ms] PMPS PBrIC_MPS ω x ω y ω z -1.0 Figure 8(a) shows the probabilities of AIS 4+ brain injury based on CSDM and MPS that increased up to about 80% after 50 msec. But the probabilities of AIS 4+ brain injury predicted by CSDM and MPS based BrIC were approximately one half of that for CSDM and MPS. During that period, the values of angular velocity around x and z axis switched from negative peak to positive peak (marked as dotted square). This result suggested that considering the values from negative (positive) peak to positive (negative) peak might contribute to improve the precision of the predictor based on CSDM and MPS. In contrast, Figure 8 (b) shows the probabilities of AIS 4+ brain injury based on CSDM, MPS and BrIC that gradually increased in accordance with the increase of angular velocities. The probabilities of AIS 4+ brain injury based on CSDM, MPS and the corresponding BrIC reached close values at 150 msec. CONCLUSIONS Vehicle crash test data for 445 ATDs obtained from NHTSA and IIHS were analyzed using SOMs and hierarchical cluster analysis to investigate the effect of time duration of angular velocities around three axes on the level of precision of BrIC. Findings are summarized below. 1. The differences of the probability of the risks between CSDM, MPS and the corresponding BrICs might be larger when the peak values of angular velocities were higher and the correspondig time durations were shorter. 2. In addition to time durations of angular velocities, incorporating the values of peak-topeak of angular velocity around each axis into the predictor s formulation might improve its level of precision. REFERENCES [1] Takhounts, E.G. et al Development of Brain Injury Criterion (BrIC), Stapp Car Crash Journal Vol. 57, pp [2] Takhounts, E.G BrIC Update: Does BrIC Depend on the Signal Time Duration? IRCOBI- NOCSAE-Snell-PDB TBI Workshop [3] Kikuchi, T Investigation of Brain Injury Mechanisms in Vehicle Crashes, In Proceedings of the JSAE Annual Congress [4] [5] [6] Kohonen, T Self-Organizing Maps Springer-Verlag Kikuchi 5

6 Nagarajan, Rangarajan Tariq, Shams GESACInc.com Tsuguhiro, Fukuda Consultant Japan Paper Number Probability model relating contact velocity and pediatric head injury severity ABSTRACT The objective of this study was to use epidemiologic, and infant cadaver drop test data to develop a probabilistic model relating probability of nondisplace skull fracture to contact velocity for infants aged up to 6-months. A secondary objective was to verify the accuracy of mass and material scaling methods used in the past to develop head injury tolerance criteria for CRABI-6M dummy. Infant fall data reported in the literature were combined with infant cadaver drop test data to develop a data set of 80 head impacts. Contact velocity for each impact in the data set was estimated from drop height; and head acceleration was estimated using pulse width from infant cadaver drop tests. Estimated peak head acceleration was related to probability of skull fracture. Estimated probability was compared with pediatric skull fracture probabilities reported in literature. The curve relating contact velocity with linear skull fracture has the form P = e ( (1.5658V c )) / (1+ e ( (1.5658V c )) ) Where V c is the contact velocity, which in this study, ranges from 1.7 m/s to 4.9 m/s. Probabilities estimated in this study agree with previously reported values thus validating the calculation procedures used in this study. INTRODUCTION Falls and motor vehicle accidents are an important cause of pediatric Emergency Department (ED) visits (Marin, et al. 2014). However, there is a lack of information about tolerance levels for various types of head injuries in infants. Traditionally, cadaver tests have been used to relate head impact injury caused by a fall. However, societal and ethical concerns have restricted pediatric cadaver testing. Limited isolated infant cadaver head testing has been conducted by Prange, (2003) and Loyd, (2011). They dropped isolated infant cadaver heads onto a rigid plate. Weber, et al. (1984, 1985) conducted full body child cadaver drop tests. They dropped uninstrumented cadavers onto rigid, and padded surfaces. All children dropped onto rigid surfaces sustained simple linear skull fractures. Two recent reports (Ruddick, et al. (2009) and Monson, et al. (2008)) discuss infant in-hospital falls. These authors reported on infant fall onto rigid hospital floors from heights ranging from 0.5m to 1.2m. Ruddick reported a number of linear skull fractures whereas Monson indicated that that only 1 of the 14 infants sustained a skull fracture. Presence or absence of skull fracture was not confirmed in all cases in both studies. Snyder et al (1963, 1977) documented falls from heights up to 11m in an attempt to estimate the relationship between injury severity, fall height, and type of contact surface. They used a combination of detailed medical and scene investigation, and computer modeling to relate fall heights and injury for free falls on to surfaces of varying stiffness. Outcomes of falls in the pediatric population has been studied either through retrospective studies of hospital admissions (Ibrahim, 2009) or using finite element models (Coats, 2003, Ibrahim, 2009, Klinich, 2002, Roth, 2008) or through dummy drop tests (Bertocchi, et al. (2003 and 2004), Coats, 2003). Li, et al. (2015) developed a finite element model to analyze falls reported by Weber (1984, 1985). They related peak head linear acceleration to probability of skull fracture. Van Ee, et al. (2009) conducted drop tests using CRABI (Child Restraint Air Bag Interaction) 6-month old dummy to reproduce Weber s drop tests. They developed a curve relating peak head acceleration to probability of pediatric skull fracture. The fall height in both these studies was set at 0.81m to match the Weber s (1964, 1985) study. Rangarajan, 1

7 Rangarajan, et al. (2013) noted that pulse width of adult cadaver head impacts is very weakly related to drop height for a given contact surface. The drop height in these tests varied from 0.6m to 2.1m. This increase of 250% in drop height caused approximately 7% decrease in the pulse width. They used this observation to calculate peak head acceleration of one child brought to the ED with a simple linear skull fracture. Infants sustain head injury from falls and in motor vehicle accidents and there is a need to evaluate probability of skull fracture from both these causes. Prior efforts have related peak head linear acceleration, which is a dependent variable, with probability of skull fracture. Rangarajan (2017) related scaled peak head acceleration of a biofidelic infant dummy head to fracture probability at three discrete fall heights (proportional to contact velocities). Li, et al (2015) and van Ee (2009) related skull fracture probability to peak head acceleration at a fixed fall height (proportional to contact velocity). To our knowledge, a continuous curve relating probability to skull fracture for various contact velocities (fall heights) is not available at present. Additionally, relations between probability of fracture and head acceleration require that tests be conducted with dummies or cadavers before such a relation can be developed. Head acceleration is a dependent variable in an impact in the sense that impact causes head acceleration. However, in many cases of falls and motor vehicle accidents, it is not too difficult a task to estimate contact velocity which is an independent variable. A probability relationship between skull fracture and an independent variable will be a very useful tool allowing researchers to develop initial estimates of skull fracture probability without having to conduct tests or to develop and exercise complicated models. In this paper, we estimate peak linear head acceleration using the procedure described by Rangarajan, et al. (2013) for infant fall cases available in literature. Literature used in this study listed the fall height, and described contact surface and consequent injuries in each fall. We then related the peak head accelerations to moderate head injuries (non-displaced skull fractures) through a probability curve. Both fracture and non-fracture cases available in literature for impacts against a rigid surface were used in our analysis. The relationship between probability of skull fracture and peak head acceleration was then converted to relationships between probability and contact velocity and / or fall height using procedures developed by Rangarajan (2013). METHODS AND MATERIALS Our objective is to develop a probabilistic relationship between contact velocity (related to fall height) and simple linear skull fracture in infants (age 6 months) for falls onto rigid surface. The process of development of the probability relationship was divided into the following steps: 1. Develop a formula relating head contact velocity and fall height. 2. Obtain Pulse Width for infant falls onto rigid surfaces from Loyd (2011). Pulse Width is defined as the difference in time between the 1 st contact of the head with the rigid surface and the beginning of the first rebound. During this period, the head deceleration goes from zero to maximum and goes back to zero. A typical head impact pulse and pulse width are shown in Fig Develop a list of infant fall cases described in literature where the falls surface was rigid and fall heights and outcome injuries were known. 4. Estimate peak head accelerations for all study cases and relate measured and estimated accelerations to probability of simple linear skull fracture. Procedure used by Rangarajan, et al. (2013, 2017) was used to estimate peak head acceleration. Details of the four steps are provided below. Calculation of head contact velocity Neglecting air friction, contact velocity V c in m/s of the head at the end of a fall of h meters under gravitation forces is given by v = 2gh (Equation 1) Where g is the gravitational constant and has a value of 9.81 m/s 2 Obtain pulse width for infant cadaver isolated head drop tests Loyd (2011) and Prange (2004) conducted a number of infant cadaver head drop tests from 0.15m and 0.3m heights. The average Pulse Width (PW) in Loyd s tests was 17 ms for forehead drops onto rigid surfaces for the age group of interest (0 age 6 Rangarajan, 2

8 months). Analysis by Rangarajan, et al. [2017] established that: midway through the pulse width. Equation (3) now reduces to: It is appropriate to use the forehead drop test pulse width for infant falls where Vertex, Occiput, and left and right parietes make first contact with a rigid surface for the velocities of interest in this study. It is appropriate to use pulse widths from isolated head drop tests for full body falls in dummies similar to Aprica 2.5 infant dummy. Pulse width average of ms calculated in Rangarajan [2017] will be used in this study. Estimate peak head acceleration We used the Impulse-Momentum theorem which is obtained by rearranging terms in Newton s second law of motion (Force = Mass * Acceleration). If force F applied to a body of mass M causes a change in velocity V during time T, then, Newton s second law can be stated as follows: F T = M V (Equation 2) Or, Impulse = Change in Momentum Substituting (Force = Mass * Acceleration), we obtain Acceleration = V/Time (Equation 3) When a head contacts a rigid surface, it starts decelerating till its velocity is zero. Deceleration reaches a maximum value when the head velocity is zero. Most damage to the head occurs during this deceleration or loading phase when head velocity goes its initial velocity to zero. To simplify calculation, we can assume that the deceleration time curve is triangular in shape. This assumption is supported by Fig. 1 which shows a reconstruction of measured head acceleration profile from one of Loyd s (2011) infant cadaver head drop tests. It is seen that during the loading phase, acceleration increases linearly from zero (at the time the head contacts the surface) to a maximum value approximately midway through the pulse width. For this shape of deceleration pulse, the average acceleration is ½ of the peak value. So, we can assume that a constant acceleration (average acceleration) is applied from the time the head contacts the surface to the end of loading phase Figure 1: Cadaver head drop test from Loyd PHA = 2 Vc/(0.5 PW) (Equation 4) PHA = h (Equation 5) Where: PHA = Peak Head Acceleration in G, m/s 2. PW = Pulse width in milliseconds, as shown in Fig. 1. From Loyd s data, we determined that PW averages to 17ms for infants (age 6-months) for 0.15m and 0.3m fall heights. h = height in meters. Obtain infant fall data from literature We used data from Loyd (2011), Monson (2009), Rangarajan (2013), Ruddick (2008), and Weber (1984, 1985) to develop a probability curve relating PHA and Probability of moderate skull fracture. Data used to develop the probability curve are summarized in Tables 1 and 2. Loyd (2011) listed drop height and measured head peak acceleration for each test. Weber (1984, 1985) dropped uninstrumented child cadavers onto rigid and non-rigid surfaces from a fixed height (81 cm) and we calculated peak head acceleration using Equation 5. Monson, et al. (2008) and Ruddick, et al. (2009) listed drop heights and we calculated peak head accelerations using Equation 5. Rangarajan, et al (2013) provided drop height and peak head acceleration (Please note that there was a calculation error in Rangarajan (2013), the corrected peak head acceleration is presented in this table). Rangarajan, 3

9 Monson (2008), Rangarajan (2013), Ruddick (2009), and Weber (1984, 1985) provided details of injuries. Ruddick (2009) reported that a number of subjects in her study were not scanned and they are not included in this analysis. Similarly, patients who were not scanned in the Monson (2008) study are not included in this analysis. In addition, one patient in the study sustained a depressed fracture. No further information was available, so this infant was not included in the analysis. Loyd (2011) conducted 5 drop tests from 2 drop heights (0.15m and 0.3m) on 7 subjects within the age group we are considering (0-6months). In these 70 (7*5*2) tests, he reported one parietal fracture of neonate (subject P12M). Peak head acceleration data from all 70 tests were used in the analysis. Table 1: Data used to develop probability of AIS2 skull fracture relationship with peak acceleration Data source Injury Description from data source # of tests L Parietal fracture. Subject P12M, m drop L No fracture. Tests with subjects 69 P03M, P05F, P06M, P07M, P08M, P12M and P13F. Ra Simple linear parietal fracture 1 Ru No clinical signs, right parietal 1 fracture Ru No clinical signs, left parietal 1 fracture Ru No clinical signs, right parietal 1 fracture, ultrasound normal. Ru Swelling in left parietal areas, left 1 parietal fracture, ultrasound normal Ru Traumatic encephalopathy, right 1 fronto-parietal fracture, cerebral contusion Ru No imaging, no clinical signs 1 Ru No imaging, no clinical signs 1 Ru No imaging, no clinical signs 1 Ru No imaging, no clinical signs 1 Ru No imaging, no clinical signs 1 Ru No imaging, no clinical signs, 1 bruise over temporal bone We Simple linear fracture 1 M No fracture, no scan 1 M No fracture, no scan 1 M No fracture, no scan 1 M No fracture, no scan 1 M No fracture, CT Scan 1 M No fracture, skull radiograph 1 M No fracture, no scan 1 Data source Injury Description from data source # of tests M No fracture, skull radiograph 1 M No fracture, no scan 1 M No fracture, skull radiograph 1 M No fracture, skull radiograph 1 M No fracture, no scan 1 M No fracture, no scan 1 M No fracture, no scan 1 Table 2: Continuation of Table 1 Data source # of tests Fall Height, m Peak head acceleratio n Estimated (E) or measured (M), G L , M L and to 112, M Ra , E Ru , E Ru , E Ru , E Ru , E Ru , E Ru , E Ru , E Ru , E Ru Ru Ru W , E M E 1.1 M E M E 1.1 M E 1.1 M E M E M M M M M M M M Keys to Column 1 (Tables 2&3): 1. L = Loyd, Rangarajan, 4

10 2. Ra = Rangarajan, 3. Ru = Ruddick, 4. M = Monson, 5. W= Weber In Tables 1 and 2, we have included all cases reported by Ruddick (2009) and Monson (2008) including infants who were not scanned. However, we included in our analysis only cases where presence or absence of skull fracture was confirmed by scans. We decided not to include Monson and Ruddick cases without scans for the following reasons: RESULTS All other cases reported included scans to confirm skull fracture. Distribution of unscanned cases indicates that roughly the same numbers, about 8, would be included in the analysis above and below the 50% probability level. Thus, the distribution would not be very different from the one where no unscanned data were included. We used data from Loyd (2011), Monson (2008), Rangarajan (2013), Ruddick (2009), and Weber (1984, 1985) to develop a probability curve for moderate skull fracture. The probability curve developed is shown in Fig. 2. Figure 2: Probability of AIS 2 head skeletal injury related to peak head acceleration The equation of the probability curve is P = e ( ( *PHA)) ( ( *PHA)) /( 1+ e (Equation 6) Where PHA = Peak Head Resultant Acceleration from Equation (5). Substituting for PHA in Equation (6) results in Equation (7) that relates contact velocity and probability of moderate skull fracture. P = e ( (6.93* (fall height)) /( 1+ e ( (6.93* (fall height)) ) (Equation 7) Estimate of injury probability obtained from this curve is similar to previously reported estimates as seen in Table 4. For easier comprehension, Table 3 below reproduces data from Fig.2. The second column shows the probability for each head acceleration listed in the first column. Columns 3 and 4 show the estimated contact velocity and estimated fall height respectively for the level of head acceleration listed in Column 1 by applying Equations 4 and 5. Table 3: Height of fall, Contact Velocity, Peak Acceleration and Skull fracture probability. Estimated Peak Head Accel, m / s 2 Skull Fracture Probability Estimated contact velocity, m/s Est. Fall Height, m Comparison of current study results with literature In the past, Mertz et al. [1986], Melvin [1995], Klinich [2003], van Ee et al. [2002], Coats [2002], and Li et al. [2015] have used various processes to relate head peak acceleration to probability of skull fracture. Results of our study are compared to the results of these studies to evaluate the Rangarajan, 5

11 appropriateness of our methodology. This comparison is presented in Tables 4 and 5. Table 4: Comparison of study estimates with literature values Sou rce Risk level of skull Fx Process Mz, 5% Scaling mass and material properties Me 5% Scaling mass and material properties K 50% Finite element model. Crash test V 50% CRABI drop tests. Vary contact surface C 50% Finite element model L 5% Finite element model. Varied contact surfaces L 50% Finite element model. Varied contact surfaces Drop Heig ht Repo r-ted N/A N/A N/A N/A N/A N/A 0.8m Stone tile, carpet 0.8m Stone tile, carpet 0.8m Stone tile, carpet Sou rce Risk level of skull Fx Process Cu 5 Algebraic formula current study Cu 50% Algebraic formula current study Table 5: Continuation of Table 4 Drop Heig ht Repo r-ted 0.15 m to 1.2m 0.15 m to 1.2m Source Acceleration Newborn 6-month old Mz, 156 Me K 85 V 82 C L L Cu 55 Cu 101 Key to Column 1 (Tables 4&5): 1. Mz = Mertz, et al (1986) 2. Me = Melvin (1995) 3. K = Klinich, et al (2003) 4. V = van Ee et al (2002) 5. C = Coats (2007) 6. L = Li et al (2015) 7. Cu = Current study Contact Surface Contact Surface Rigid steel plate Rigid Steel plate The probability curve in Fig. 2 and data in Tables 4 and 5 indicate that our estimates are close those provided by other researchers. However, previous probability curves (Li, (2015), Van Ee (2009)) have been constructed with data from one fall height. Our probability curve is designed to handle all fall heights up to 1.2m and directly relates level of injury to fall height and contact velocity. To the best of our knowledge, this is the only effort, apart from that of Snyder (1977) to relate fall heights to injury severities. Subjects in Snyder s study generally fell more than 3m and the youngest subject was 13 month old. Rangarajan, 6

12 CONCLUSIONS 1. A probability curve constructed using a mixture of estimated and measured peak head accelerations for falls less than or equal to 1.2 m is comparable with those constructed using complex finite element models and dummy drop tests. Thus, it seems feasible to use the proposed algebraic formula to estimate skull fracture probability for contact velocities less than 4.3 m/s. 2. Finite element models yield very detailed information about the fracture, and response of the brain that is not provided by the proposed algebraic model. However, finite element models of child head require a large amount of data to define the geometry of the head, and material properties of the brain, skull, scalp, and sutures. It has been hard to generate these data given restrictions in child cadaver and child cadaver tissue testing. 3. Instrumented dummies are generally expensive and testing with them requires expensive ancillary equipment such as data acquisition hardware and software. Dummies also require periodic calibration using specialized equipment. So, both dummy testing and finite element models require more effort than most busy medical centers can afford to invest. We are hopeful that the simple analytic procedure discussed in this paper will encourage researchers to collect data from a large number of fall cases that come to the ED. Since the proposed model requires only 2 pieces of information from patients fall height and a detailed list of injuries on the first visit to ED, we are hopeful that more researchers will collect information about a lot of falls thus making the proposed model more robust. Such a robust model can be used by finite element modelers to refine their models and conduct in-depth investigations into the effect of falls. 4. Linear acceleration of the CG of the head which is the output of the proposed model is related arithmetically to angular acceleration and angular velocity of the head by the formula: Head angular acceleration = Head linear acceleration / radius of rotation. Center of rotation in infants is not known but it has been estimated to be around C2, or about 1/4 the length of the neck. This simple formula can be used to relate intracranial injuries to angular acceleration in a large number of cases thus forming the basis of an investigation into the effect of impact on MTBI and TBI. 5. This work verifies the appropriateness of material and geometric scaling techniques proposed by Melvin (1995) 6. The analytic method used in this study can be expanded to older children and used to design better pedestrian head impact protection for children. 7. These results provide guidance for the development of test devices to model child head impact and a reduction in the need for child cadaver tests. Limitations 1. The proposed probability curve does not separate effects of falls onto various parts of the head. However, analysis presented in Rangarajan (2017) indicates All impacts used in this study are against rigid surfaces. There is a need to extend this work to other surfaces such as seat back cushions, carpets, soil, etc. 2. The procedure used assumes that deceleration of the rest of the body does not significantly affect peak linear deceleration of the infant head. While this is true of the Aprica 2.5 dummy (Rangarajan, et al (2017)), it may not be applicable to live infants. References 1. Bertocci, GE, Pierce, MC, Deemer, E, Aguel, F, Janosky, JE, Vogeley, E. Using Test Dummy Experiements to Investigate Pediatric Injury Risk in Simulated Short-Distance Falls. Arch Pediatr Adolesc Med. May;157(5): Bertocci, GE, Pierce MC, Deemer E, Aguel F, Janosky JE, Vogeley E.. Influence of fall height and impact surface on biomechanics of feet-first free falls in children. Injury Apr;35(4): Coats, B. Mechanics of head impacts in infants. PhD Thesis, University of Pennsylvania Ibrahim, N.G. et al. Influence of age and fall type on head injuries n infants and toddlers. Int. J. Dev Neurosci. May: 30(3): Klinich K, Hublert G, Schneider L. Estimating infant head injury criteria and impact response using crash reconstruction and finite element modeling. Stapp Car Crash J. 46: Li, Z, Liu, W, Zhang, J, Hu, J. Prediction of skull fracture injury risk for children 0-9 months old through validated parametric finite element model and cadaver test reconstruction. Rangarajan, 7

13 Int J Legal Med. Sep;129 (5): Loyd, AM. Studies of the Human Head from Neonate to Adult: An Inertial, Geometrical and Structural Analysis with Comparisons to the ATD Head. PhD Thesis. Duke University Marin, JR, Weaver, MD, Yealy, DM, Mannix, RC. Trends in Visits for Traumatic Brain Injury to Emergency Departments in the United States. JAMA.May 14;311(18): Melvin, J. Injury Assessment Reference Values for the CRABI 6-Month Infant Dummy in a Rear-Facing Infant Restraint with Airbag Deployment. SAE International Congress and Exposition Proceedings. SAE Paper No Mertz, H. A Procedure of Normalizing Impact Response Data. SAE Technical Paper Society of Automotive Engineers, Warrendale, PA Mertz, HJ. Injury assessment values used to evaluate Hybrid III response measurements. NHTSA Docket Notice Mertz, H., Irwin, A., Melvin, J., Stalnakar, RI, Beebe, MS. Size, Weight and Biomechanical Impact Response Requirements for Adult Size Small Female and Large Male Dummies. In automotive frontal impacts. SP = 782, , SAE Warrendale, PA Monson, SA. In-hospital Falls of Newborn Infants: Data from a Multihospital Heath Care System. P Pediatrics. Aug;122(2):e Prange, MT, Luck, JF, Dibb, A, Van Ee, CA, Nightingale, RW, Myers, BS. Mechanical properties and anthropometry of the human infant head. Stapp Car Crash J. Nov;48: Rangarajan, N, DeRosia, J, Humm, J, Tomas, D, Cox,J. An improved method to calculate paediatric skull fracture threshold. Enhanced Safety Vehicle Conference Proceedings, Seoul, South Korea Rangarajan, N, Shams, T, Jenny, C, and Fukuda, T. Probability of pediatric skull fracture at various contact velocities. Enhanced Safety Vehicle Conference Proceedings, Detroit Ruddick, C, Ward, Martin PWP, and De San Lazaro, C Head trauma outcomes of verifiable falls in newborn babies. Arch Dis Child Fetal Neonatal Ed Mar;95(2):F Snyder, R. G Human Survivability of Extreme Impacts in Free-Fall. Report, Civil Aeromedical Institute, Oklahoma City Snyder, RG, Fousts, DR, and Bowman, BM Study of Impact Tolerance Through Free-Fall Investigations. Report UM-HSRI- 77-8, University of Michigan Van Ee, C, Moroski-Brown, B, Raymond, D, Thibault, K, Hardy, W, Plunkett, J. Evaluation and refinement of the CRABI-6 anthropometric test device injury criteria for skull fracture. Proceedings of the ASME 2009 International Mechanical Engineering Congress & Exposition, Lake Buena Vista Weber, W. [Experimental studies of skull fractures in infants]. Z Rechtsmed. 92(2): (Ger). 22. Weber, W. [Biomechanical fragility of the infant skull]. Z Rechtsmed. ;94(2): (Ger). Rangarajan, 8

14 RECONSTRUCTION OF A SIDE IMPACT ACCIDENT WITH FAR-SIDE OCCUPANT USING HBM DISCUSSION OF POTENTIAL APPLICATION OF VIRTUAL HBM WITHIN A FAR-SIDE OCCUPANT PROTECTION ASSESSMENT Christian Mayer Jan Dobberstein Uwe Nagel Daimler AG Germany Ravikiran Chitteti Ghosh Pronoy Sammed Pandharkar Mercedes-Benz Research and Development India Pvt. Ltd India Paper Number ABSTRACT Advanced Human Body FE models are now being used extensively in the development process of vehicle safety systems. This tool on one hand aids in the optimization of restraint systems and on the other hand also provides a detailed analysis of injury mechanisms when used within accident reconstruction. A good documented (injury patterns & physical loading conditions) real world crash and its reconstruction not only ensure further development of vehicle safety, but also allows further improvement of these Human Body Models in terms of biomechanical validity and injury prediction capability. This is particularly important, as injury prediction should not only be based on physical thresholds or isolated tissue based injury parameters but should also allow a population based probabilistic estimation of injury risk. Therefore the main objective of this study was the reconstruction and detailed analysis of a real world side crash using a numerical HBM. This real world side crash was chosen from the DBCars in-house accident database of Daimler. In the selected case, a medium sized Mercedes car was struck at approximately the front wheel on the passenger side and had a rollover subsequently. The driver sustained mainly abdominal injuries. A THUMS V4 male model was used to represent the driver of the struck car and to reconstruct the injuries. The probabilistic injury criteria for pelvis fracture, recently published by J. Peres et al. and the probabilistic rib fracture criteria published by J. Forman et al. were implemented to the post-processing tool DYNASAUR. Further stress/strain based injury predictors for other body regions were also used within this study. The real world crash and the injury patterns of the driver were compared and discussed with statistical data whether it can be considered as representative for a typical far-side load case. Finally the applicability of Virtual Testing and use of a HBM within an assessment protocol are discussed for this far-side load case.

15 INTRODUCTION In March 2015, Euro NCAP published a Strategic Road Map, identifying major domains that are supposed to focus on key real life crash scenarios and that are supposed to be addressed by new and updated safety technology [1]. One of these domains are side impacts, since according to Euro NCAP far-side impacts have been overlooked in testing and seem equally important as casualties from near side impact. Suppliers of restraint systems aim to address these upcoming requirements with new restraint systems ( center bags ) [2]. On the other hand, data analysis based on NASS/CDS and CIREN done by Brumbelow et. al. [3] highlights the variety of impact conditions occurring in far-side impacts. This study analyzes impact direction of force, impact area, number of collisions, impact severity etc. in relevant cases, and investigates injured body regions. Given this detailed understanding, Brumbelow assigns a low priority to far-side scenarios contrary to Euro NCAP and conclude that they are difficult to address with a single test configuration. Accident data analysis of GIDAS data [4] resulted in n=43 belted front row car driver or occupants of cars registered 2000 or later who were seated in the farside position of a side collision by cars or other vehicles. Only 67% of the vehicles showed a compartment impact (Fig. 1, top left). Less than 30% of these had a perpendicular impact similar to side impact scenarios currently standard in laboratory tests (Fig. 1, top right). Injuries due to head-to-head contacts are not a major issue in the field, since head injuries in far-side cases with close occupant do not predominate. For both groups, thorax injuries are high. The share of abdominal injuries is also comparably high, many of them not further specified traumatic AIS2 injuries to liver or kidney (Fig. 1, bottom). Several studies and researches have already shown, that human body finite element (FE) models are the method of choice to reconstruct real world crash injury outcome and demonstrated in the same way, that injury predictors and criteria used with these tools could be further validated and consolidated in comparison with real world crash scenarios. Golman et al. [5] discussed the use of a HBM in detail within an accident reconstruction of a nearside crash selected from the CIREN (Crash Injury Research and Engineerin Network) database. He also comprehensively analysed the HBM response and injury prediction capability by applying several injury metrics from literature or recently developed and implemented to the human body model which has been used in his study. 1,8% Far Side w. close occ. Far Side alone x% had AIS2+ injury in body region of all belted front 2,6% 2,6% occupants in far side impacts 2,1% 0,5% 0,6% 0,5% 0,2% 0,0% 0,0% AIS2+ Head AIS2+ Thorax AIS2+ Abdomen AIS2+ Neck AIS2+ Pelvis Figure 1. Accident statistics Far-side This study now aims to: 1. Understand in more detail the injury mechanisms in far-side collisions and to show the capabilities and hurdles of human body simulations of these scenarios. 2. Apply a post-processing for evaluating injury risk. 3. Apply aspects of Virtual Testing in conducting accident reconstruction. Therefore, similar to the study of Golman also the Total Human for Safety (THUMS) [6] [7] was used to represent the occupant, now in a far-side configuration, in this real case reconstruction. In addition, respectively in contrast to previous studies, a post processing tool was applied with THUMS to evaluate the HBM response and injury risk. This addresses mainly the aspect of standardization and harmonization of injury risk prediction and evaluation by virtual human body models and finally the applicability within assessment protocols. Mayer 2

16 This topic, respectively more general Virtual Testing, was extensively discussed within the European FP7 research project IMVITER [8]. The consortium comprised of 15 partners from Germany, France, Italy, Spain, Hungary and Greece and represented the main actors involved in EC motor vehicle type approval process. In the project, Virtual Testing was considered as use of simuation models in the assessment procedures of regulatory acts, replacing real tests or supporting real tests in terms of supplemental assessment procedure. Therefore, the following definition was formulated: Virtual Testing (VT) can be defined as the assessment of any kind of requirement imposed on a physical part or system, which is conventionally accomplished through some kind of test, but performed using a numerical model instead. Thus, VT inherently replaces tests (also named Real Testing - RT) by simulation models and test results by simulation predictions. Beside the demonstration of such a numerical assessment in four pilot cases, from which one focused on the application of HBM, also a generic VT type approval implementation process was developed. This process introduces the Verfification, Validation (V&V) and finally the type approval assessment in three consecutive phases (Fig. 2). The European resarch project SafeEV (Safe Small Electric Vehicles through Advanced Simulation Methodologies) [9] was initated as follow-up project within the 7 th European Framework Programme and consolidated the main findings from IMVITER. Mainly FE-HBM were implemented to a proposed assessment procedures and applied to evaluate advanced safety solutions for pedestrians and car occupants. Finally relevant and crucial processing steps within a projected tool chain were identified and discussed in the course of the project e.g. V&V (Verification & Validation), recommendations for comparability of codes and especially harmonized and standardized post-processing methods including criteria definition were mainly identified as Best Practice respectively Key Building Blocks for VT and the implementation of HBM (Fig.3). Figure 3. SafeEV tool chain of a virtual assessment including application of HBM and related Key Building Blocks Therefore the purpose of this study was also to make use of some of the findings from these European projects and discuss e.g. the capability of a postprocessing tool and recently published criteria for THUMS V4 to predict the injuries in a real world case on the one hand and to comment on their possible applicability within assessment procedures on the other hand. METHOD A real world side crash was chosen from the DBCars in-house accident database of Daimler. The Mercedes accident research unit investigates and reconstructs severe accidents of Mercedes-Benz cars since Figure 2. General IMVITER VT implementation flowchart The accident was reconstructed by PC-Crash [10]. The PC-Crash software is one of the leading tools for traffic accident reconstruction. Collisions up to 32 vehicles can be simulated in 2D and also in 3D. Car to car accidents, car to motorcycles, car to pedestrian Mayer 3

17 accidents, occupant movement and also roll over can be calculated. Several databases of all common cars and motorcycles are included in PC-Crash. The PC crash simulation was taken as a basis for the FE simulation analysis carried out in this study. THUMS V4.02 AM50 occupant Human Body Model was used as the driver of A-Class which was involved in the accident. This HBM was used to assess and predict the injuries happened in actual crash scenario. The output data from the THUMS V4 simulations were mainly processed using the tool DYNASAUR V0.03 (PYTHON based - DYNASAUR was developed by Graz University of Technology and a first application was demonstrated within the EU research project SafeEV [11] [12] [13]. With specified input concerning criteria and injury predictors, the tool runs automatically the complete assessment process and creates a standardized report. The tool is flexible in terms of possible adaption to different HBM and also additional implementation of criteria. The current version used within this study includes evaluation schemes for head injuries, rib fractures, organ damage, pelvis injuries, bone fractures and ligament ruptures. Far-side Accident Reconstruction Accident Scenario A 2015 Mercedes-Benz A-Class crossed a junction and was hit at the right side by a 2014 Mercedes E-Class station wagon. The impact was at approx. 80 angle at the right front wheel of the A-Class (see Fig. 4 & 6). The A-Class spun and rolled onto the passenger side where it came to the final position. There was an activation of the belt tensioner, driver and passenger front bag, knee bag, passenger side bag and right window bag. The driver of the A-Class was the only occupant (44years old, 178 cm, 125 kg). He was belted and suffered the following injuries: right kidney contusion, left shoulder contusion, left hip contusion, left hand laceration, right hand contusion, right lower leg contusion, whiplash of the cervical spine of the neck. The abdominal injury of the farside driver results most likely from a contact with the center console during the impact from the E-Class. It is also very likely, that the whiplash and the right lower leg contusion occurred during this impact phase. In contrast it might be resonably assumed that the other injuries of the driver occurred during the rollover in the second phase of this accident. Accident Reconstruction by PC-Crash Based on the accident investigation, different parameters like point of collision, speeds of the cars and friction were changed until the calculated final position corresponds as good as possible to the real final position. PC-Crash output provided results in terms of collision parameters and kinematics of the vehicles, which were now used in the FE reconstruction and simulation of the structural interaction of the vehicles. The final, reconstructed accident configuration and kinematics by PC-Crash is shown in figure 5-7. Figure 5. Sketch of accident scene blue: impacting vehicle (E-Class), black: struck vehicle (A-Class) Figure 4. Struck vehicle Mercedes A-Class: point of impact at the right front wheel Figure 6. Impact configuration Mayer 4

18 detailed front fascia was absent in the A-Class model. The missing details primarily do not provide any structural strength, which is provided by front bumper cross member present in the model. Figure 7. Movement of the cars from impact to the final position A vehicle-to-vehicle impact simulation was performed, with position and velocity inputs from PC-Crash and accident reconstruction data, to validate the FE model setup. FE Vehicle and Interior Models Accident reconstruction by PC-Crash gives a good visual impression of the accident. However, the amount of data that could be extracted from PC-Crash simulation is limited. Also, in the resultant velocity output from PC-Crash, as shown in Fig. 8, a sharp drop of velocity was observed, which would result in unrealistic acceleration levels in the next stage of the reconstruction. Thus, a need for full vehicle-to- vehicle FE crash simulation was identified. Figure 9. Full Vehicle FE models The observed deformation pattern, however, was different from that in real crash. In addition, the second impact of the two vehicles was not achieved with the initial setup. Figure 8. Resultant velocity v/s time output from PC-Crash Explicit FE simulations with LS-DYNA are capable of providing realistic representation of the vehicle deformations as well as occupant kinematics under accident scenario. Fig. 9 shows the full vehicle models used. The FE models include detailed BIW parts, engine compartment and the details of the components inside engine compartment, doors with trims, wheels and the suspension assembly etc. A To validate the FE model setup, a parametric study was performed with impact locations (distance between center of front axle of the two vehicles), impact angles and friction between tire and road. Table 1 lists the parameters and their respective ranges studied. Parameter Impact location Impact angle Coefficient of friction Value Range Final value studied 0 mm-1400 mm 1300 mm Table 1. Parameters of study Mayer 5

19 The setup shown in Fig. 10 was observed to match the deformation patterns on the vehicles. The second impact was also achieved as shown in Fig.11. The time difference between the two impacts was 239ms. A time difference of approximately 238ms was achieved in FE simulation. Fig. 12 shows photographs of vehicle taken after the accident with deformations in FE simulation overlaid on top of them. The full vehicle simulation performed was observed to be computationally expensive. To reduce the computational time required and reduce the complexity of the study, a sled model of A-Class car was created from the full vehicle model. Fig.13 shows a so-called sled model, in which the components in vicinity of occupant e.g. seat, center console, door trims, steering wheel, and instrument panel were modelled as flexible and the rest including the structure were modelled as rigid. Figure 10. Final position of models for the full vehicle simulation Figure 13. A-Class sled model Figure 11. Impacts of the two cars in full vehicle simulation To transfer the motion from full vehicle crash simulation to the sled model, three nodes were identified where minimal deformation was observed. Displacement data from these three nodes were extracted and applied to the respective nodes in the sled to impart motion to the sled model. The sled simulation was overlaid over full vehicle simulation to verify transfer of displacement data (Fig. 14). Figure 14. Sled model simulation overlaid over full vehicle simulation Figure 12. Overlaid deformation patterns from simulation Mayer 6

20 Human Body Model The driver (far-side configuration) in this numerical study was represented by a THUMS Version %-ile male occupant model (AM50). THUMS was and is still developed by Toyota Motor Corporation and Toyota Central R&D Labs. Version 4.02 was released in 2015 [6] [7] with a total number of approximately 1.9 million elements and about nodes. In this version, the inner organs are modelled in detail. The model represents an average adult with a standing height of 178.6cm and a weight of 77.6kg. It should be noted, that height of the driver matches well with 50 th percentile male height, but the weight of the driver is much higher than that of THUMS V4. The possible obese body of the driver is currently not represented in this study. This fact is discussed in the FE result and discussion section. THUMS V4 model was positioned in A-Class sled model on the driver side based on ergonomics posture calculations. Seat squashing was done to ensure proper contact between HBM and seat foam. A 3-point seatbelt was routed around the HBM using Primer 12.1 software. The seatbelt was equipped with pre-tensioner system. The positioned model is shown in Fig.15. It was observed during crashed vehicle inspection that the driver front airbag and knee airbag were deployed. Figure 15. THUMS V4 positioned in A-Class sled model Injury Metrics The THUMS V4 is a detailed human body model and has capability to predict injuries in complex loading conditions. In this study, injuries in ribs, pelvis and kidney were analyzed and compared with those observed in the accident. From the injury data recorded in the actual accident, AIS 1 injuries i.e. whiplash, hand contusions and lacerations, shoulder contusion were not analyzed in this study. The injury criteria used to assess the injuries are shown in Table 2. Body organ Ribs Pelvis Kidney Criteria developed by A.H. Burstein et. al.(1976) [17] J. Foreman et. al. (2012) [15] J. Peres et. al. (2016) [18] K. Shigeta et. al. (2009) [19] J. Snedecker et al. (2005) [20] Description of the criteria Deterministic criteria. 3% plastic strain in cortical bones as fracture limit Probabilistic criteria. Based upon maximum local strains Probabilistic criteria. Based upon maximum principal strain in pelvic bone. Deterministic criteria based upon maximum principal strains in soft organs. Criteria based upon the local Strain Energy Density (SED) at time of rupture in kidney. Local SED of 43 kj/mm 3 was set as limit. Table 2. Injury prediction criteria Mayer 7

21 DYNASAUR tool, referred in the earlier section, was used to post-process the results. It is a PYTHON based post-processing tool developed to predict injuries based upon injury prediction criteria available in literature. The tool can be configured by the user to incorporate new injury criteria. The injury criteria which are available in the tool and which were added as new are shown in Table 3. Injury criteria available in DYNASAUR CSDM SUFEHM Head injury Long bone fracture Ribs (Foreman) Internal organs - Heart - Interstine - Spleen - Lung - Liver Newly added injury criteria in DYNASAUR Internal organs - Kidney Neck (for SUFEHNM) Pelvis (Peres) Figure 16. Fringe plot of plastic strains in ribs. 10 th rib shows more than 3% plastic strain Ribs - Forman Criteria Fig. 17 shows the probability of rib fracture as indicated by DYNASAUR tool. It was noted from the output that the right ribs have significantly higher probability of fracture than the left ribs, as right side of torso was coming in contact with center console. The 10th rib on the right side of THUMS V4 was showing almost 100% probability of fracture, similar to the prediction by Burstein criteria. Table 3. Injury criteria in DYNASAUR Additional criteria were evaluated using traditional post-processing tools such as LS- PrePost. RESULTS FE Simulations Occupant & Injury Prediction The oblique nature of this far-side impact causes the HBM to have a predominant higher lateral component of movement than frontal. The first impact of the two vehicles was most severe resulting in the abdomen and pelvis of the HBM colliding with center console. The HBM was analyzed for resulting injuries based upon the injury metrics discussed in the previous section. Ribs - Burstein Criteria Cortical part of 10 th rib was observed to have more than 3% plastic strain, predicting fracture. Fig. 16 shows the plastic strain plot of the ribs. Figure 17. Probability of rib fracture from DYNASAUR tool (Forman Criteria) Pelvis Fig. 18 shows the probability of injury in pelvic bone as per the criteria developed by J. Peres et al. The output, from DYNASAUR tool, showed more than 95% probability of AIS 2+ injury happening in pelvic bone. Mayer 8

22 Figure 18. Probability of AIS2+ pelvic injury Kidney - Shigeta criteria For analyzing the injury in kidney, maximum principal strain of 50% was maintained as a limit for injury. Out of 100% solid elements in right kidney, 63% of elements were observed to fall above the limit compared to 5% of elements out of 100% elements in the left kidney. Fig. 19 shows the number of elements in kidney plotted against the percentage of the Maximum Principal Strain limit at the time of HBM making contact with the center console. Figure 19. Distribution of number of elements in the kidneys with respect to the percentage of Maximum Principal Strain limit Kidney - Snedecker criteria Strain Energy Density (SED) in the kidney was analyzed for injury prediction. It was noted, that the right kidney exceeded the minimum rupture limit, 43 kj/m3, given by Snedecker. The maximum strain energy density observed in the kidney is kj/m3, indicating injury. The left kidney, however, did not exceed the minimum rupture limit. Fig. 20 shows the strain energy density fringe plot in the kidneys. These observations establish the explanation of higher probability of rupture in right kidney. Figure 20. SED fringe plots for kidneys DISCUSSION Rib cage injury Thorax injuries account for one of the highest number of AIS2+ injuries in far-side accident scenarios as discussed in the introduction section. It might be reasonably assumed that the main cause of the thoracic injuries is the contact with the center console of the car. As shown in Fig. 21, the contact between center console and thorax resulted in significant deformation of the ribcage, causing the lower ribcage to bend laterally at around 98ms. The rib injury prediction (Burstein and Forman criteria) showed high probability of fracture in the 10 th rib. Whereas, the real occupant did not endure any injuries to ribcage. This difference could be attributed to the anthropometric difference (weight / obesity) between the THUMS V4 and the real occupant. Figure 21. Deformation and plastic strain in Pelvic injury The prediction by DYNASAUR tool showed a high risk of AIS 2+ pelvic injuries. The criteria, being Mayer 9

23 based upon local maximum principal strains, is sensitive towards the modelling related local strain concentrations, as reported by Peres et. al [18]. From the plastic strain plot in Fig. 22, it was observed that the high strains were only occurring at the hip joint area and the localized strains might over-predict the injuries in the pelvic bone. Figure 24. Effective stress v/s time in kidney solid elements. Figure 22. Plastic strains in pelvic bone Kidney Injury As mentioned in the result section, the right kidney was found to be more susceptible to injury than the left. The difference between the injury levels could be explained as the organs on the right (impact side) were subjected to direct loading from the vehicle interiors unlike the organs on the left (non-impact side). Because of the difference in the loading, the right kidney was observed to have higher internal energy (Fig. 23). The right kidney attains maximum internal energy of 3.13 J at 100 ms whereas left kidney has maximum internal energy of 2.06 J at 96ms. Figure 23. Kidney Internal energy v/s time (red: right kidney, blue: left kidney) Subsequently, the elements in the right kidney also show high stress values as depicted in Fig. 24 and predicting higher probability for injury. LIMITATIONS OF THE STUDY The deformations of the vehicles from photographs helped reconstruct the accident scenario. However, no data on the kinematics of occupant in real event is available. Therefore, the estimation of the contacts of occupant with vehicle interior with respect to accident event timeline and subsequent injuries is purely based on the FE Simulations. These predicted injuries are compared with the injury data from the accident data. The posture of the occupant inside the car as well as the seat position at the time of accident is also not known. The best possible occupant posture and seat position was computed using the ergonomics data and taken as input. The THUMS V4 model used in this study is a western 50 th percentile male model. The weight of HBM is around 77.6 kg and the model height is around 178.6cm. The body height of the driver of A-Class in accident scenario matches quite well with the model, whereas the weight (125kg) of the occupant differs significantly from the model weight. This difference could have led to different kinematics of the occupant and subsequently different injury risk. Therefore, the prediction of the rib fracture in the numerical study was interpreted as a first approximation to reality. Finally the FE simulation just focused on the first phase of the accident scenario. The rollover was not taken into account for the injury risk estimation respectively occupant kinematics and further contacts with interior parts. For this, it was assumed that the rollover was less critical in terms Mayer 10