This article was downloaded by: [University of Sussex Library] On: 17 August 2015, At: 07:48 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: 5 Howick Place, London, SW1P 1WG
Traffic Injury Prevention Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/gcpi20
Finite Element Model Prediction of Pulmonary Contusion in Vehicle-to-Vehicle Simulations of RealWorld Crashes ab
Kerry A. Danelson a
& Joel D. Stitzel
ab
Wake Forest University School of Medicine, Winston-Salem, North Carolina
b
Virginia Tech, Wake Forest University School of Biomedical Engineering and Sciences, Winston-Salem, North Carolina Accepted author version posted online: 08 Jan 2015.
Click for updates To cite this article: Kerry A. Danelson & Joel D. Stitzel (2015) Finite Element Model Prediction of Pulmonary Contusion in Vehicle-to-Vehicle Simulations of Real-World Crashes, Traffic Injury Prevention, 16:6, 627-636, DOI: 10.1080/15389588.2014.995266 To link to this article: http://dx.doi.org/10.1080/15389588.2014.995266
PLEASE SCROLL DOWN FOR ARTICLE Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http:// www.tandfonline.com/page/terms-and-conditions
Traffic Injury Prevention (2015) 16, 627–636 C Taylor & Francis Group, LLC Copyright ISSN: 1538-9588 print / 1538-957X online DOI: 10.1080/15389588.2014.995266
Finite Element Model Prediction of Pulmonary Contusion in Vehicle-to-Vehicle Simulations of Real-World Crashes KERRY A. DANELSON1,2 and JOEL D. STITZEL1,2 1 2
Wake Forest University School of Medicine, Winston-Salem, North Carolina Virginia Tech, Wake Forest University School of Biomedical Engineering and Sciences, Winston-Salem, North Carolina
Traffic Injury Prevention 2015.16:627-636.
Received 15 May 2014, Accepted 2 December 2014
Objective: Pulmonary contusion (PC) is a common chest injury following motor vehicle crash (MVC). Because this injury has an inflammatory component, studying PC in living subjects is essential. Medical and vehicle data from the Crash Injury Research and Engineering Network (CIREN) database were utilized to examine pulmonary contusion in case occupants with known crash parameters. Method: The selected CIREN cases were simulated with vehicle finite element models (FEMs) with the Total HUman Model for Safety (THUMS) version 4 as the occupant. To match the CIREN crash parameters, vehicle simulations were iteratively improved to optimize maximum crush location and depth. Fifteen cases were successfully modeled with the simulated maximum crush matching the CIREN crush to within 10%. Following the simulations, stress and strain metrics for the elements within the lungs were calculated. These injury metrics were compared to patient imaging data to determine the best finite element predictor of pulmonary contusion. Results: When the thresholds were evaluated using volumetric criteria, first principal strain was the metric with the least variation in the FEM prediction of PC. Conclusions: A preliminary threshold for maximum crush was calculated to predict a clinically significant volume of pulmonary contusion. Keywords: lung, injury, pulmonary contusion, CIREN, computer modeling
Introduction Pulmonary contusion (PC) is a common injury following motor vehicle crashes (MVCs) and blunt chest trauma (Shorr et al. 1987) with potentially long-term respiratory deficiencies following injury (Kishikawa et al. 1991). Previous NASS studies have shown that PC accounts for over 35% of Abbreviated Injury Scale 3+ chest injuries (Stitzel et al. 2010). Side impact MVCs account for a disproportionate number of PC cases (NHTSA 2006). Gayzik et al. (2009) found that frontal crashes occur approximately twice as often as side impacts. When only the occupants who have PC are examined, the crash mode is nearly split, with 51% of PC cases resulting from frontal impacts and 48% from side impacts. PC volume thresholds have been correlated to a significantly higher risk of developing acute respiratory distress syndrome (ARDS);
Managing Editor David Viano oversaw the review of this article. Address correspondence to Joel D. Stitzel, PhD, Center for Injury Biomechanics, Wake Forest University, 575 Patterson Ave, Suite 120, Winston-Salem, NC 27101. E-mail: [email protected] Color versions of one or more of the figures in the article can be found online at www.tandfonline.com/gcpi.
therefore, the percentage volume of injured lung has clinical implications (Becher et al. 2012; Miller et al. 2001). Experimental and computational models of PC in various animals have been used in previous studies for MVC and blast injury (Bass et al. 2008; Gayzik et al. 2011; Hoth et al. 2006; Stuhmiller et al. 1988). These studies have quantified insult and outcome data due to the similar physiological response in the animal. There have been studies of PC in humans; however, they have investigated the relationship between PC and clinical complications (Becher et al. 2012; Miller et al. 2001). These human studies do not examine the mechanical insult that led to the development of PC. The intent of this study was to evaluate PC in humans who have been exposed to a quantified mechanical insult from a MVC by using the Crash Injury Research and Engineering Network (CIREN) database. CIREN cases include a reconstruction of the crash by a trained crash investigator and a case review by medical professionals, crash experts, and biomechanical engineers. The database contains detailed crash information, medical imaging, injury causation scenarios, and occupant outcomes (NHTSA 2010; Schneider et al. 2011). Previous simulations to determine human injury mechanisms by using occupied vehicle-to-vehicle crashes are limited. Siegel et al. (2010) and Belwadi et al. (2012) have investigated aortic injury in MVC using CIREN cases. The simulations
628
Traffic Injury Prevention 2015.16:627-636.
were then evaluated for strain and pressure within the aorta. One limitation of these studies was that they used an older version of a 50th percentile male human body model (HBM) without the anatomic detail available in current models. Additionally, no variations due to occupant size were implemented. To address the size limitation, the current study will implement volumetric scaling. However, the simulation methodology builds on this existing work. The current study expands on previous studies of animal PC and simulated crash reconstruction to create a novel injury metric to evaluate PC in human subjects. The hypothesis of this study was that an evidence-based volumetric strain metric for PC injury threshold in a HBM can be established by using CIREN data to understand the mechanical insult to the thorax and relate this known insult to a computed tomography (CT) measured volume of pulmonary contusion. To answer this hypothesis, several questions were asked to guide the study analysis: 1. Can CIREN cases be simulated given the data available in the database? 2. Given occupant imaging data, how are the simulation results correlated to the occupant outcome to evaluate volume of PC using an HBM? 3. Will the high strain elements in the lung qualitatively be in the same location as the identified occupant PC? 4. Using the work from the previous questions, can a simple linear regression between crash parameters and PC volume be established to understand the relationship between the insult and the outcome?
Methods Case Selection Cases simulated were near-side impact crash configurations due to the high incidence of PC following this crash configuration (NHTSA 2006). Cases with any tree impact or a rollover were discarded because these events involved a complex interaction between the vehicle and the ground/tree that would be difficult to simulate. The resulting data set consisted of 15 cases. The volume, location, and extent of pulmonary contusion were quantified in these occupants in a previous study by using the CIREN database CT scans (Danelson et al. 2011). For these occupants, both high attenuation lung, defined as any area of increased radio-opacity within the lung, and PC were identified. High attenuation lung was segmented using a semi-automated method based on Hounsfield unit (HU) values. Healthy tissue had a default range of −1,024 and −562 HU and high attenuation was defined as HU values higher than −562. Given the selected high attenuation lung volume, a board-certified radiologist examined the occupant’s CT images and selected the areas that were isolated PC. High attenuation lung not selected as PC was primarily atelectasis or aspiration. The other data from the CIREN database used for simulation preparation were occupant height and weight, vehicle change in velocity, vehicle maximum
Danelson and Stitzel crush, maximum crush location, and principal direction of force (PDOF).
Finite Element Models The finite element solver used for this study was LS-DYNA (MPP, Version 971, LSTC, Livermore, CA) run on a computer cluster. The vehicle model selected for the simulations was a 2001 Ford Taurus. This FE vehicle model was available with a full seat, dash panel, and steering column from the National Crash Analysis Center (NCAC 2008, 2012). The remainder of the NCAC vehicle models did not have this level of detail for the interior components. Because the Taurus was the struck vehicle, only CIREN cases with a sedan as the case vehicle were simulated. The striking vehicle was a sedan (Ford Taurus 2001 FEM), SUV (Ford Explorer 2007 FEM), or truck (Chevrolet Silverado 2007 FEM). The vehicle models used for this study were validated by the NCAC by simulating frontal crash configurations from the NHTSA database (NCAC 2012). The occupant model selected for this work was the Total HUman Model for Safety (THUMS, Toyota Central R&D Labs, Nagakute, Japan) version 4 due to its enhanced biofidelity over previous versions of the THUMS (Shingeta et al. 2009). Specifically, THUMS version 4 has internal organs with lungs and the heart located in the thoracic cavity (Shingeta et al. 2009). THUMS was validated by comparison to published postmortem human subject tests (Iwamoto et al. 2002, 2003; Shingeta et al. 2009). The regional thoracic response of the model was compared to frontal impactor testing (Kroell et al. 1971, 1974; Toyota Motor Corporation 2010) and to belt loading tests (Cesari and Bouquet 1990). The individual THUMS thoracic organ response was validated by comparing simulation results, specifically when strain and pressure exceeded set injury levels, to recorded organ injury following autopsy in the Kroell impact tests (Kroell et al. 1971, 1974; Shingeta et al. 2009). The lung elements were modeled with ∗ MAT ELASTIC FLUID and tetrahedral elements. The element section definition was solid with an element formulation of 13, a one-point nodal pressure tetrahedron (Hallman et al. 2011). In the THUMS manual, a pressure threshold for pulmonary contusion was suggested. This threshold was based on work conducted by Schaefer et al. (1958) on canine subjects to determine lung injury threshold during diving decompression (Toyota Motor Corporation 2010). THUMS was validated for side impact loading by simulating lateral thoracic (Shaw et al. 2006) and pelvic (Guillemot et al. 1997) impacts. The lateral thoracic impact assessed the force deflection of the thorax following lateral rib impact. In the pelvis, a rigid sphere ball impacted the acetabulum. These lateral loading conditions evaluated the regional response of the THUMS; however, the model is currently lacking a full-body side impact validation.
Occupant and Vehicle Scaling Differences between the base FEMs and simulated vehicle or occupant models were addressed through model scaling. Only the THUMS 50th male percentile model was used; therefore, male and female occupants were simulated with the
Pulmonary Contusion
629
Table 1. Occupant anthropometry comparison between CIREN and THUMS Height (cm) Male/ Weight (kg) Sim female # CIREN CIREN FEM CIREN
Traffic Injury Prevention 2015.16:627-636.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Male Male Female Female Female Female Male Female Female Male Female Female Female Male Male
93 99 50 68 64 61 48 52 93 88 47 68 64 63 75
88 74 56 59 59 59 63 63 63 74 63 56 63 88 88
183 180 160 162 163 163 165 168 170 180 165 157 168 185 188
Occupant Length Scale 1.06 1.00 0.90 0.93 0.93 0.93 0.90 0.95 0.95 1.00 0.95 0.90 0.95 1.06 1.06
BMI (kg/m2)
PC Volume CIREN FEM (%) 28 31 20 26 24 23 18 18 32 27 17 28 23 18 21
28 25 22 23 23 23 24 24 24 25 24 22 24 28 28
5.42 0.77 7.85 1.03 3.13 1.91 21.23
4.96 6.17
male model. The occupants were grouped by comparing the occupant height to the nearest 25th percentile individual to facilitate length scaling (Gordon et al. 1989). The percentiles selected for scaling were 25th, 50th, and 75th percentile female and a 50th (original THUMS) and 95th percentile male by height (Table 1). During this process, the body mass index (BMI) of each occupant was determined and average BMI values were calculated for the size groups. Cases where the occupant was more than 2 standard deviations from the average BMI of the group were not simulated. The scaling method selected was length scaling the occupant size evenly along the X, Y , and Z axes in the LS-DYNA input deck. No separate adjustments were made for occupant mass; however, there were mass changes due to the size changes. The weight difference between the CIREN occupant and the scaled THUMS model had an average value of 12.7 kg with a standard deviation of 8.6 kg (Table 1). Vehicle models were mass scaled (without the occupant mass) in the LS-DYNA input deck so the masses of simulated vehicles matched those of the recorded CIREN database vehicle masses (Appendix 1, Table A1, see online supplement). By using the input deck mass scaling, the location of the vehicle center of gravity was maintained. Occupant Positioning The original location of THUMS relative to the vehicle driver seat was forward and rotated about the Z-axis. Prior to settling, the THUMS model was translated and rotated using the LS-PrePost interface and the LS-DYNA input deck until it was approximately 2 mm above the driver seat cushions. THUMS was settled into the seat with a 500-ms gravity simulation. The seat frame was positioned along the seat track to avoid occupant knee bolster penetration and approximate the recorded CIREN occupant seat position. Following settling, belt fitting was completed using the LS-PrePost interface (Figure 1). Belt material properties were provided by a restraint manufacturer as representative of an average vehicle belt. Re-
Fig. 1. (A) Seated 50th percentile male occupant with restraints fitted and (B) at full door engagement during impact.
gardless of the actual vehicle configuration, the belt configuration was the same for all cases, with rigid attachment points, no load limiters, and no pretensioners. All cases simulated did not have side airbag deployment. Occupied Vehicle-to-Vehicle Simulations All cases were simulated for 200 ms to capture the full occupant loading phase of the event. The full simulations were iteratively improved to optimize the maximum crush. Maximum crush and the location of crush were evaluated and used to iteratively adjust the simulation because these parameters were identified as the best predictors of high attenuation lung and PC volume in a previous study (Danelson et al. 2011). The striking vehicle
Traffic Injury Prevention 2015.16:627-636.
630
Danelson and Stitzel
Fig. 2. Crush as a function of time (top) with vehicle cross sections to illustrate the vehicle deformation and occupant response at maximum crush (bottom left) and the simulation end time point (bottom right).
impact location and initial velocity were varied for this optimization. The initial striking vehicle location along the length of the struck car was based on the vehicle crush photographs and the crush profile from the CIREN database. The angle of the impact was simulated using the PDOF entered in the database. Each impacting vehicle was given an initial velocity equal to the impact delta V . The maximum crush location was measured relative to the front or rear axis of the vehicle in CIREN. This location was then mapped to a specific node in the simulation and used for future measurements of maximum crush. Once the first simulation was completed, the maximum crush and the delta V were determined for the simulation. To calculate maximum crush, a plane was created using 3 points defined on a frame member on the sill opposite the striking vehicle that did not deform during the simulation. From this plane, the perpendicular distance from the node at the maximum crush location to the plane was calculated for each simulation time point. To compare the simulation crush to the CIREN database, the crush was measured at the end of the simulation. The vehicle door structure used a ∗ MAT PIECEWISE LINEAR PLASTICITY material model; therefore, the final measured crush value was less than dynamic crush measured during the impact (Figure 2; NCAC 2012). In this example, the maximum dynamic crush was 37.2 cm compared to 33.6 cm at the end of the simulation (0.2 s). Following the initial iterative simulations, the location of the maximum crush on the simulated struck vehicle was compared to the specified crush location node. Initially, gross adjustments of the location of the striking vehicle along the struck vehicle were necessary to match the maximum crush location between the case and the simulation. Once the crush location matched, the delta V of the striking vehicle was ad-
Fig. 3. Steps in the iterative improvement process to match the maximum crush between the simulation and the CIREN case.
justed to change the crush depth until the simulation static maximum crush matched to within 10% of the CIREN maximum crush value (Figure 3). The delta V was evaluated for the simulations but it was not used as an optimization parameter. The side crush profile on the CIREN cases and the simulation crush profiles were qualitatively a close match (Figure 4; all cases are shown in Appendix 2, see online supplement).
Fig. 4. (A) Comparison of the CIREN case vehicle and (B) the simulated vehicle crush.
Pulmonary Contusion
631
Traffic Injury Prevention 2015.16:627-636.
Simulation Postprocessing Following the simulation, the maximum values of the following stress- and strain-based metrics were evaluated for each element in the lung over the full course of the simulation: first principal strain, shear strain, octahedral shear stress, principal stress, and triaxial strain. Additionally, strain rates and the products of strain and strain rate were calculated. The threshold for each metric, by case occupant, was determined by identifying the metric value necessary to match the simulation and the CIREN occupant percentage volume PC. First, the FEM strain values in all lung elements were ordered from highest to lowest. Plots of the cumulative volume over the full range of the injury metric values were created to illustrate the relationship of strain to volume of affected lung (Figure 5). For each strain level, the cumulative volume of the lung that these values represented was calculated. Therefore, the left limit of the plot (lowest PC volume) correlates to the highest strain value in the lung and the right limit (maximum PC volume) represents the lowest strain value. Additionally, a qualitative assessment of the location of high strain elements in the lung was completed.
Statistical Analysis Given the calculated strain thresholds based on occupant PC volume, a statistical analysis was performed on each of the selected metrics (first principal strain, first principal strain rate, etc.). This analysis included mean strain threshold that predicted the correct volume of PC, the resulting standard deviation of the strain threshold values, a minimum and maximum value, and coefficient of variation (COV). The strain metric with the lowest COV, when considering all simulations, was selected to predict the volume of PC. The COV was used because a low value indicated a metric that was robust enough to predict injury given a range of boundary conditions. Therefore, the selected metric would be least dependent on the crash parameters. Using Figure 5 as an example, first principal strain had a tighter grouping of lines in the plot and a lower COV. To evaluate the validity of this assumption, correlation testing was performed to determine whether the threshold was independent of the crash parameters. This testing would indicate a consistent threshold between subjects exposed to different crash loadings. To assess the correlation, the calculated thresholds were plotted against delta V , lateral delta V , and maximum crush. Then, a linear regression was performed and analyzed for statistical significance using the t statistic, α = .05, and goodness of fit from the R2 value. A future goal of this work would be to develop simplified relationships between measured crash parameters, such as maximum crush and delta V , to predict PC volume without conducting full simulations; therefore, an initial analysis of this type was performed. Given an average threshold value calculated from the case occupant results, PC volume was determined for each occupant. The volume of predicted PC was calculated for each simulation by selecting elements exceeding the specified strain threshold at any time during the simulation. The predicted volume was plotted as a function of measured simulation crush, total measured simulation delta
Fig. 5. Cumulative volume plots for the first principal strain (top) and first principal strain times strain rate (bottom) element magnitudes.
V , and lateral simulation delta V . Then linear and exponential regressions were calculated to determine whether there was a relationship between the predicted simulation volume of PC and crash metrics. The accuracy of these regressions was assessed by calculating statistical significance using the t statistic, α = .05, and R2 values.
Results CIREN cases were successfully simulated using NCAC vehicle models and THUMS given the data available in the CIREN database. The 15 cases selected for simulation did not have error terminations in the final simulation configurations. Of the 15 cases, 9 cases had isolated sections of PC, as evaluated by a board-certified radiologist. These 9 cases were used in the analysis and selection of the best strain metric threshold for
632
Danelson and Stitzel
Table 2. Simulation matrix with the resulting delta V calculated from the vehicle center of gravity accelerometer compared to the CIREN case WinSMASH-calculated delta V Lateral delta V Simulation type
Traffic Injury Prevention 2015.16:627-636.
Taurus vs. Taurus Taurus vs. Taurus SUV vs. Taurus SUV vs. Taurus SUV vs. Taurus SUV vs. Taurus SUV vs. Taurus SUV vs. Taurus SUV vs. Taurus SUV vs. Taurus SUV vs. Taurus Truck vs. Taurus Truck vs. Taurus Truck vs. Taurus Truck vs. Taurus
Total delta V
CIREN (kph)
Simulation (kph)
Difference (kph)
CIREN (kph)
Simulation (kph)
18 18 41 27 24 50 36 40 52 23 35 36 33 29 52
26.80 22.21 33.96 26.16 33.08 50.54 35.58 34.80 47.31 57.32 30.56 31.93 26.18 17.64 37.48
8.80 4.21 7.04 0.84 9.08 0.54 0.42 5.20 4.69 34.32 4.44 4.07 6.82 11.36 14.52
28 21 42 27 24 51 37 43 55 23 37 38 35 29 52
36 22 38 27 33 51 36 35 50 62 31 33 26 18 38
Lateral deviation difference summary Mean (kph) Standard deviation (kph) Coefficient of variation Minimum value (kph) Maximum value (kph)
7.76 8.35 1.08 0.42 34.32
these simulations. The simulation maximum static crush (sim max crush) was compared to the database maximum static crush (max crush) values (Appendix 1, Table A2, see online supplement) and all simulation cases were within 10% of the database measured maximum crush. Qualitatively, the high strain elements in the FEM were in a similar location compared to the CIREN occupant identified PC (Appendix 3, see online supplement). For the majority of the case occupants’ high attenuation tissue, there were isolated sections of tissue within the lung as well as a more diffuse section of high attenuation tissue in the posterior aspect of the lung (Figure 6A) in the semi-automated segmentations. The PC identified by the radiologist was a smaller portion of the total high attenuation results (Figure 6B; Danelson et al. 2011). In the simulation results, there were isolated regions of the highest strain elements, based on PC volume, in the anterior and lateral aspects of the lung without diffuse element selection in the posterior lung (Figure 6C). These locations were generally comparable to the case occupant PC. Maximum principal stress, maximum shear stress, octahedral shear stress, and triaxial mean strain were excluded from further consideration after the qualitative strain analysis because the tissue selected using these metrics was diffusely spread throughout the lung. The remaining metrics were first principal strain, maximum shear strain, first principal strain rate, maximum shear rate, the product of first principal strain and rate, and the product of shear strain and rate (Appendix 1, Table A3, see online supplement). The crash parameter to threshold correlation testing resulted in no statistically significant correlations or regression fits; therefore, the calculated thresholds were independent of the crash parameters. The calculated R2 values ranged from
Traffic Injury Prevention Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/gcpi20
Finite Element Model Prediction of Pulmonary Contusion in Vehicle-to-Vehicle Simulations of RealWorld Crashes ab
Kerry A. Danelson a
& Joel D. Stitzel
ab
Wake Forest University School of Medicine, Winston-Salem, North Carolina
b
Virginia Tech, Wake Forest University School of Biomedical Engineering and Sciences, Winston-Salem, North Carolina Accepted author version posted online: 08 Jan 2015.
Click for updates To cite this article: Kerry A. Danelson & Joel D. Stitzel (2015) Finite Element Model Prediction of Pulmonary Contusion in Vehicle-to-Vehicle Simulations of Real-World Crashes, Traffic Injury Prevention, 16:6, 627-636, DOI: 10.1080/15389588.2014.995266 To link to this article: http://dx.doi.org/10.1080/15389588.2014.995266
PLEASE SCROLL DOWN FOR ARTICLE Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http:// www.tandfonline.com/page/terms-and-conditions
Traffic Injury Prevention (2015) 16, 627–636 C Taylor & Francis Group, LLC Copyright ISSN: 1538-9588 print / 1538-957X online DOI: 10.1080/15389588.2014.995266
Finite Element Model Prediction of Pulmonary Contusion in Vehicle-to-Vehicle Simulations of Real-World Crashes KERRY A. DANELSON1,2 and JOEL D. STITZEL1,2 1 2
Wake Forest University School of Medicine, Winston-Salem, North Carolina Virginia Tech, Wake Forest University School of Biomedical Engineering and Sciences, Winston-Salem, North Carolina
Traffic Injury Prevention 2015.16:627-636.
Received 15 May 2014, Accepted 2 December 2014
Objective: Pulmonary contusion (PC) is a common chest injury following motor vehicle crash (MVC). Because this injury has an inflammatory component, studying PC in living subjects is essential. Medical and vehicle data from the Crash Injury Research and Engineering Network (CIREN) database were utilized to examine pulmonary contusion in case occupants with known crash parameters. Method: The selected CIREN cases were simulated with vehicle finite element models (FEMs) with the Total HUman Model for Safety (THUMS) version 4 as the occupant. To match the CIREN crash parameters, vehicle simulations were iteratively improved to optimize maximum crush location and depth. Fifteen cases were successfully modeled with the simulated maximum crush matching the CIREN crush to within 10%. Following the simulations, stress and strain metrics for the elements within the lungs were calculated. These injury metrics were compared to patient imaging data to determine the best finite element predictor of pulmonary contusion. Results: When the thresholds were evaluated using volumetric criteria, first principal strain was the metric with the least variation in the FEM prediction of PC. Conclusions: A preliminary threshold for maximum crush was calculated to predict a clinically significant volume of pulmonary contusion. Keywords: lung, injury, pulmonary contusion, CIREN, computer modeling
Introduction Pulmonary contusion (PC) is a common injury following motor vehicle crashes (MVCs) and blunt chest trauma (Shorr et al. 1987) with potentially long-term respiratory deficiencies following injury (Kishikawa et al. 1991). Previous NASS studies have shown that PC accounts for over 35% of Abbreviated Injury Scale 3+ chest injuries (Stitzel et al. 2010). Side impact MVCs account for a disproportionate number of PC cases (NHTSA 2006). Gayzik et al. (2009) found that frontal crashes occur approximately twice as often as side impacts. When only the occupants who have PC are examined, the crash mode is nearly split, with 51% of PC cases resulting from frontal impacts and 48% from side impacts. PC volume thresholds have been correlated to a significantly higher risk of developing acute respiratory distress syndrome (ARDS);
Managing Editor David Viano oversaw the review of this article. Address correspondence to Joel D. Stitzel, PhD, Center for Injury Biomechanics, Wake Forest University, 575 Patterson Ave, Suite 120, Winston-Salem, NC 27101. E-mail: [email protected] Color versions of one or more of the figures in the article can be found online at www.tandfonline.com/gcpi.
therefore, the percentage volume of injured lung has clinical implications (Becher et al. 2012; Miller et al. 2001). Experimental and computational models of PC in various animals have been used in previous studies for MVC and blast injury (Bass et al. 2008; Gayzik et al. 2011; Hoth et al. 2006; Stuhmiller et al. 1988). These studies have quantified insult and outcome data due to the similar physiological response in the animal. There have been studies of PC in humans; however, they have investigated the relationship between PC and clinical complications (Becher et al. 2012; Miller et al. 2001). These human studies do not examine the mechanical insult that led to the development of PC. The intent of this study was to evaluate PC in humans who have been exposed to a quantified mechanical insult from a MVC by using the Crash Injury Research and Engineering Network (CIREN) database. CIREN cases include a reconstruction of the crash by a trained crash investigator and a case review by medical professionals, crash experts, and biomechanical engineers. The database contains detailed crash information, medical imaging, injury causation scenarios, and occupant outcomes (NHTSA 2010; Schneider et al. 2011). Previous simulations to determine human injury mechanisms by using occupied vehicle-to-vehicle crashes are limited. Siegel et al. (2010) and Belwadi et al. (2012) have investigated aortic injury in MVC using CIREN cases. The simulations
628
Traffic Injury Prevention 2015.16:627-636.
were then evaluated for strain and pressure within the aorta. One limitation of these studies was that they used an older version of a 50th percentile male human body model (HBM) without the anatomic detail available in current models. Additionally, no variations due to occupant size were implemented. To address the size limitation, the current study will implement volumetric scaling. However, the simulation methodology builds on this existing work. The current study expands on previous studies of animal PC and simulated crash reconstruction to create a novel injury metric to evaluate PC in human subjects. The hypothesis of this study was that an evidence-based volumetric strain metric for PC injury threshold in a HBM can be established by using CIREN data to understand the mechanical insult to the thorax and relate this known insult to a computed tomography (CT) measured volume of pulmonary contusion. To answer this hypothesis, several questions were asked to guide the study analysis: 1. Can CIREN cases be simulated given the data available in the database? 2. Given occupant imaging data, how are the simulation results correlated to the occupant outcome to evaluate volume of PC using an HBM? 3. Will the high strain elements in the lung qualitatively be in the same location as the identified occupant PC? 4. Using the work from the previous questions, can a simple linear regression between crash parameters and PC volume be established to understand the relationship between the insult and the outcome?
Methods Case Selection Cases simulated were near-side impact crash configurations due to the high incidence of PC following this crash configuration (NHTSA 2006). Cases with any tree impact or a rollover were discarded because these events involved a complex interaction between the vehicle and the ground/tree that would be difficult to simulate. The resulting data set consisted of 15 cases. The volume, location, and extent of pulmonary contusion were quantified in these occupants in a previous study by using the CIREN database CT scans (Danelson et al. 2011). For these occupants, both high attenuation lung, defined as any area of increased radio-opacity within the lung, and PC were identified. High attenuation lung was segmented using a semi-automated method based on Hounsfield unit (HU) values. Healthy tissue had a default range of −1,024 and −562 HU and high attenuation was defined as HU values higher than −562. Given the selected high attenuation lung volume, a board-certified radiologist examined the occupant’s CT images and selected the areas that were isolated PC. High attenuation lung not selected as PC was primarily atelectasis or aspiration. The other data from the CIREN database used for simulation preparation were occupant height and weight, vehicle change in velocity, vehicle maximum
Danelson and Stitzel crush, maximum crush location, and principal direction of force (PDOF).
Finite Element Models The finite element solver used for this study was LS-DYNA (MPP, Version 971, LSTC, Livermore, CA) run on a computer cluster. The vehicle model selected for the simulations was a 2001 Ford Taurus. This FE vehicle model was available with a full seat, dash panel, and steering column from the National Crash Analysis Center (NCAC 2008, 2012). The remainder of the NCAC vehicle models did not have this level of detail for the interior components. Because the Taurus was the struck vehicle, only CIREN cases with a sedan as the case vehicle were simulated. The striking vehicle was a sedan (Ford Taurus 2001 FEM), SUV (Ford Explorer 2007 FEM), or truck (Chevrolet Silverado 2007 FEM). The vehicle models used for this study were validated by the NCAC by simulating frontal crash configurations from the NHTSA database (NCAC 2012). The occupant model selected for this work was the Total HUman Model for Safety (THUMS, Toyota Central R&D Labs, Nagakute, Japan) version 4 due to its enhanced biofidelity over previous versions of the THUMS (Shingeta et al. 2009). Specifically, THUMS version 4 has internal organs with lungs and the heart located in the thoracic cavity (Shingeta et al. 2009). THUMS was validated by comparison to published postmortem human subject tests (Iwamoto et al. 2002, 2003; Shingeta et al. 2009). The regional thoracic response of the model was compared to frontal impactor testing (Kroell et al. 1971, 1974; Toyota Motor Corporation 2010) and to belt loading tests (Cesari and Bouquet 1990). The individual THUMS thoracic organ response was validated by comparing simulation results, specifically when strain and pressure exceeded set injury levels, to recorded organ injury following autopsy in the Kroell impact tests (Kroell et al. 1971, 1974; Shingeta et al. 2009). The lung elements were modeled with ∗ MAT ELASTIC FLUID and tetrahedral elements. The element section definition was solid with an element formulation of 13, a one-point nodal pressure tetrahedron (Hallman et al. 2011). In the THUMS manual, a pressure threshold for pulmonary contusion was suggested. This threshold was based on work conducted by Schaefer et al. (1958) on canine subjects to determine lung injury threshold during diving decompression (Toyota Motor Corporation 2010). THUMS was validated for side impact loading by simulating lateral thoracic (Shaw et al. 2006) and pelvic (Guillemot et al. 1997) impacts. The lateral thoracic impact assessed the force deflection of the thorax following lateral rib impact. In the pelvis, a rigid sphere ball impacted the acetabulum. These lateral loading conditions evaluated the regional response of the THUMS; however, the model is currently lacking a full-body side impact validation.
Occupant and Vehicle Scaling Differences between the base FEMs and simulated vehicle or occupant models were addressed through model scaling. Only the THUMS 50th male percentile model was used; therefore, male and female occupants were simulated with the
Pulmonary Contusion
629
Table 1. Occupant anthropometry comparison between CIREN and THUMS Height (cm) Male/ Weight (kg) Sim female # CIREN CIREN FEM CIREN
Traffic Injury Prevention 2015.16:627-636.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Male Male Female Female Female Female Male Female Female Male Female Female Female Male Male
93 99 50 68 64 61 48 52 93 88 47 68 64 63 75
88 74 56 59 59 59 63 63 63 74 63 56 63 88 88
183 180 160 162 163 163 165 168 170 180 165 157 168 185 188
Occupant Length Scale 1.06 1.00 0.90 0.93 0.93 0.93 0.90 0.95 0.95 1.00 0.95 0.90 0.95 1.06 1.06
BMI (kg/m2)
PC Volume CIREN FEM (%) 28 31 20 26 24 23 18 18 32 27 17 28 23 18 21
28 25 22 23 23 23 24 24 24 25 24 22 24 28 28
5.42 0.77 7.85 1.03 3.13 1.91 21.23
4.96 6.17
male model. The occupants were grouped by comparing the occupant height to the nearest 25th percentile individual to facilitate length scaling (Gordon et al. 1989). The percentiles selected for scaling were 25th, 50th, and 75th percentile female and a 50th (original THUMS) and 95th percentile male by height (Table 1). During this process, the body mass index (BMI) of each occupant was determined and average BMI values were calculated for the size groups. Cases where the occupant was more than 2 standard deviations from the average BMI of the group were not simulated. The scaling method selected was length scaling the occupant size evenly along the X, Y , and Z axes in the LS-DYNA input deck. No separate adjustments were made for occupant mass; however, there were mass changes due to the size changes. The weight difference between the CIREN occupant and the scaled THUMS model had an average value of 12.7 kg with a standard deviation of 8.6 kg (Table 1). Vehicle models were mass scaled (without the occupant mass) in the LS-DYNA input deck so the masses of simulated vehicles matched those of the recorded CIREN database vehicle masses (Appendix 1, Table A1, see online supplement). By using the input deck mass scaling, the location of the vehicle center of gravity was maintained. Occupant Positioning The original location of THUMS relative to the vehicle driver seat was forward and rotated about the Z-axis. Prior to settling, the THUMS model was translated and rotated using the LS-PrePost interface and the LS-DYNA input deck until it was approximately 2 mm above the driver seat cushions. THUMS was settled into the seat with a 500-ms gravity simulation. The seat frame was positioned along the seat track to avoid occupant knee bolster penetration and approximate the recorded CIREN occupant seat position. Following settling, belt fitting was completed using the LS-PrePost interface (Figure 1). Belt material properties were provided by a restraint manufacturer as representative of an average vehicle belt. Re-
Fig. 1. (A) Seated 50th percentile male occupant with restraints fitted and (B) at full door engagement during impact.
gardless of the actual vehicle configuration, the belt configuration was the same for all cases, with rigid attachment points, no load limiters, and no pretensioners. All cases simulated did not have side airbag deployment. Occupied Vehicle-to-Vehicle Simulations All cases were simulated for 200 ms to capture the full occupant loading phase of the event. The full simulations were iteratively improved to optimize the maximum crush. Maximum crush and the location of crush were evaluated and used to iteratively adjust the simulation because these parameters were identified as the best predictors of high attenuation lung and PC volume in a previous study (Danelson et al. 2011). The striking vehicle
Traffic Injury Prevention 2015.16:627-636.
630
Danelson and Stitzel
Fig. 2. Crush as a function of time (top) with vehicle cross sections to illustrate the vehicle deformation and occupant response at maximum crush (bottom left) and the simulation end time point (bottom right).
impact location and initial velocity were varied for this optimization. The initial striking vehicle location along the length of the struck car was based on the vehicle crush photographs and the crush profile from the CIREN database. The angle of the impact was simulated using the PDOF entered in the database. Each impacting vehicle was given an initial velocity equal to the impact delta V . The maximum crush location was measured relative to the front or rear axis of the vehicle in CIREN. This location was then mapped to a specific node in the simulation and used for future measurements of maximum crush. Once the first simulation was completed, the maximum crush and the delta V were determined for the simulation. To calculate maximum crush, a plane was created using 3 points defined on a frame member on the sill opposite the striking vehicle that did not deform during the simulation. From this plane, the perpendicular distance from the node at the maximum crush location to the plane was calculated for each simulation time point. To compare the simulation crush to the CIREN database, the crush was measured at the end of the simulation. The vehicle door structure used a ∗ MAT PIECEWISE LINEAR PLASTICITY material model; therefore, the final measured crush value was less than dynamic crush measured during the impact (Figure 2; NCAC 2012). In this example, the maximum dynamic crush was 37.2 cm compared to 33.6 cm at the end of the simulation (0.2 s). Following the initial iterative simulations, the location of the maximum crush on the simulated struck vehicle was compared to the specified crush location node. Initially, gross adjustments of the location of the striking vehicle along the struck vehicle were necessary to match the maximum crush location between the case and the simulation. Once the crush location matched, the delta V of the striking vehicle was ad-
Fig. 3. Steps in the iterative improvement process to match the maximum crush between the simulation and the CIREN case.
justed to change the crush depth until the simulation static maximum crush matched to within 10% of the CIREN maximum crush value (Figure 3). The delta V was evaluated for the simulations but it was not used as an optimization parameter. The side crush profile on the CIREN cases and the simulation crush profiles were qualitatively a close match (Figure 4; all cases are shown in Appendix 2, see online supplement).
Fig. 4. (A) Comparison of the CIREN case vehicle and (B) the simulated vehicle crush.
Pulmonary Contusion
631
Traffic Injury Prevention 2015.16:627-636.
Simulation Postprocessing Following the simulation, the maximum values of the following stress- and strain-based metrics were evaluated for each element in the lung over the full course of the simulation: first principal strain, shear strain, octahedral shear stress, principal stress, and triaxial strain. Additionally, strain rates and the products of strain and strain rate were calculated. The threshold for each metric, by case occupant, was determined by identifying the metric value necessary to match the simulation and the CIREN occupant percentage volume PC. First, the FEM strain values in all lung elements were ordered from highest to lowest. Plots of the cumulative volume over the full range of the injury metric values were created to illustrate the relationship of strain to volume of affected lung (Figure 5). For each strain level, the cumulative volume of the lung that these values represented was calculated. Therefore, the left limit of the plot (lowest PC volume) correlates to the highest strain value in the lung and the right limit (maximum PC volume) represents the lowest strain value. Additionally, a qualitative assessment of the location of high strain elements in the lung was completed.
Statistical Analysis Given the calculated strain thresholds based on occupant PC volume, a statistical analysis was performed on each of the selected metrics (first principal strain, first principal strain rate, etc.). This analysis included mean strain threshold that predicted the correct volume of PC, the resulting standard deviation of the strain threshold values, a minimum and maximum value, and coefficient of variation (COV). The strain metric with the lowest COV, when considering all simulations, was selected to predict the volume of PC. The COV was used because a low value indicated a metric that was robust enough to predict injury given a range of boundary conditions. Therefore, the selected metric would be least dependent on the crash parameters. Using Figure 5 as an example, first principal strain had a tighter grouping of lines in the plot and a lower COV. To evaluate the validity of this assumption, correlation testing was performed to determine whether the threshold was independent of the crash parameters. This testing would indicate a consistent threshold between subjects exposed to different crash loadings. To assess the correlation, the calculated thresholds were plotted against delta V , lateral delta V , and maximum crush. Then, a linear regression was performed and analyzed for statistical significance using the t statistic, α = .05, and goodness of fit from the R2 value. A future goal of this work would be to develop simplified relationships between measured crash parameters, such as maximum crush and delta V , to predict PC volume without conducting full simulations; therefore, an initial analysis of this type was performed. Given an average threshold value calculated from the case occupant results, PC volume was determined for each occupant. The volume of predicted PC was calculated for each simulation by selecting elements exceeding the specified strain threshold at any time during the simulation. The predicted volume was plotted as a function of measured simulation crush, total measured simulation delta
Fig. 5. Cumulative volume plots for the first principal strain (top) and first principal strain times strain rate (bottom) element magnitudes.
V , and lateral simulation delta V . Then linear and exponential regressions were calculated to determine whether there was a relationship between the predicted simulation volume of PC and crash metrics. The accuracy of these regressions was assessed by calculating statistical significance using the t statistic, α = .05, and R2 values.
Results CIREN cases were successfully simulated using NCAC vehicle models and THUMS given the data available in the CIREN database. The 15 cases selected for simulation did not have error terminations in the final simulation configurations. Of the 15 cases, 9 cases had isolated sections of PC, as evaluated by a board-certified radiologist. These 9 cases were used in the analysis and selection of the best strain metric threshold for
632
Danelson and Stitzel
Table 2. Simulation matrix with the resulting delta V calculated from the vehicle center of gravity accelerometer compared to the CIREN case WinSMASH-calculated delta V Lateral delta V Simulation type
Traffic Injury Prevention 2015.16:627-636.
Taurus vs. Taurus Taurus vs. Taurus SUV vs. Taurus SUV vs. Taurus SUV vs. Taurus SUV vs. Taurus SUV vs. Taurus SUV vs. Taurus SUV vs. Taurus SUV vs. Taurus SUV vs. Taurus Truck vs. Taurus Truck vs. Taurus Truck vs. Taurus Truck vs. Taurus
Total delta V
CIREN (kph)
Simulation (kph)
Difference (kph)
CIREN (kph)
Simulation (kph)
18 18 41 27 24 50 36 40 52 23 35 36 33 29 52
26.80 22.21 33.96 26.16 33.08 50.54 35.58 34.80 47.31 57.32 30.56 31.93 26.18 17.64 37.48
8.80 4.21 7.04 0.84 9.08 0.54 0.42 5.20 4.69 34.32 4.44 4.07 6.82 11.36 14.52
28 21 42 27 24 51 37 43 55 23 37 38 35 29 52
36 22 38 27 33 51 36 35 50 62 31 33 26 18 38
Lateral deviation difference summary Mean (kph) Standard deviation (kph) Coefficient of variation Minimum value (kph) Maximum value (kph)
7.76 8.35 1.08 0.42 34.32
these simulations. The simulation maximum static crush (sim max crush) was compared to the database maximum static crush (max crush) values (Appendix 1, Table A2, see online supplement) and all simulation cases were within 10% of the database measured maximum crush. Qualitatively, the high strain elements in the FEM were in a similar location compared to the CIREN occupant identified PC (Appendix 3, see online supplement). For the majority of the case occupants’ high attenuation tissue, there were isolated sections of tissue within the lung as well as a more diffuse section of high attenuation tissue in the posterior aspect of the lung (Figure 6A) in the semi-automated segmentations. The PC identified by the radiologist was a smaller portion of the total high attenuation results (Figure 6B; Danelson et al. 2011). In the simulation results, there were isolated regions of the highest strain elements, based on PC volume, in the anterior and lateral aspects of the lung without diffuse element selection in the posterior lung (Figure 6C). These locations were generally comparable to the case occupant PC. Maximum principal stress, maximum shear stress, octahedral shear stress, and triaxial mean strain were excluded from further consideration after the qualitative strain analysis because the tissue selected using these metrics was diffusely spread throughout the lung. The remaining metrics were first principal strain, maximum shear strain, first principal strain rate, maximum shear rate, the product of first principal strain and rate, and the product of shear strain and rate (Appendix 1, Table A3, see online supplement). The crash parameter to threshold correlation testing resulted in no statistically significant correlations or regression fits; therefore, the calculated thresholds were independent of the crash parameters. The calculated R2 values ranged from
