Embed Size (px)
Citation preview
Foi
XMa
b
c
d
a
ARR2A
KSHDSFCA
1
aSttmAotat
f
0h
Accident Analysis and Prevention 55 (2013) 232– 241
Contents lists available at SciVerse ScienceDirect
Accident Analysis and Prevention
j ourna l h om epage: www.elsev ier .com/ locate /aap
inite element analysis of occupant head injuries: Parametric effectsf the side curtain airbag deployment interaction with a dummy headn a side impact crash
ingqiao Denga,∗, S. Potulab,c, H. Grewalb,c, K.N. Solankid,.A. Tschoppb,c, M.F. Horstemeyerb,c
School of Mechanical Engineering and Automation, Xihua University, Chengdu, Sichuan 610039, ChinaCenter for Advanced Vehicular Systems, Mississippi State University, MS 39726, USADepartment of Mechanical Engineering, Mississippi State University, MS, USASchool of Engineering of Matter, Transport and Energy, Arizona State University, Tempe, AZ 85287, USA
r t i c l e i n f o
rticle history:eceived 20 January 2012eceived in revised form4 September 2012ccepted 13 March 2013
eywords:ide curtain airbagead injury criterionesign of experimentside impact
a b s t r a c t
In this study, we investigated and assessed the dependence of dummy head injury mitigation on the sidecurtain airbag and occupant distance under a side impact of a Dodge Neon. Full-scale finite element vehiclesimulations of a Dodge Neon with a side curtain airbag were performed to simulate the side impact. Owingto the wide range of parameters, an optimal matrix of finite element calculations was generated usingthe design method of experiments (DOE); the DOE method was performed to independently screen thefinite element results and yield the desired parametric influences as outputs. Also, analysis of variance(ANOVA) techniques were used to analyze the finite element results data. The results clearly show thatthe influence of moving deformable barrier (MDB) strike velocity was the strongest influence parameteron both cases for the head injury criteria (HIC36) and the peak head acceleration, followed by the initialairbag inlet temperature. Interestingly, the initial airbag inlet temperature was only a ∼30% smaller
inite element analysisrashNOVA
influence than the MDB velocity; also, the trigger time was a ∼54% smaller influence than the MDBvelocity when considering the peak head accelerations. Considering the wide range in MDB velocitiesused in this study, results of the study present an opportunity for design optimization using the differentparameters to help mitigate occupant injury. As such, the initial airbag inlet temperature, the triggertime, and the airbag pressure should be incorporated into vehicular design process when optimizing forthe head injury criteria.
. Introduction
Improving occupant safety remains among the most importantnd challenging design considerations in the automotive industry.tatistics from the National High-Way Traffic Safety Administra-ions (NHTSA) shows that five million vehicle crashes occurred inhe United States in the year 2009, which injured more than two
illion people and claimed more than 33,808 lives (NHTSA, 2009).mong the injured or killed, head injury is the most common causef death during worst-case side impact scenarios, and the head is
he most likely injured body region, even among occupants usingthree-point restraint system. Also, prior analyses demonstratehat some occupant restraint devices, such as airbags and seatbelts,
∗ Corresponding author. Tel.: +1 662 497 4404/+86 13980726249;ax: +1 662 325 5433.
E-mail addresses: dxq [email protected], [email protected] (X. Deng).
001-4575/$ – see front matter © 2013 Elsevier Ltd. All rights reserved.ttp://dx.doi.org/10.1016/j.aap.2013.03.016
© 2013 Elsevier Ltd. All rights reserved.
contribute to reduce the risk of head injury and enhance occu-pant safety during impact crashes (Gabauer and Gabler, 2010). Theyare typically installed as thorax-or torso-only airbags, torso-headairbags (combination airbags), or separate torso and head airbags(curtain or inflatable tubular structures). Although federal regula-tions do not exist in the United Stated for side airbags, such systemsare gaining popularity, because of public awareness for safety andtheir potential injury mitigation characteristics (Fujiyama et al.,2011; Dix et al., 2011).
Nonetheless, there is still a controversy about serious occu-pant injuries caused by airbag deployment, mainly attributed tothe airbag interacting with the occupant position. Moreover, asthe number of cars equipped with airbags increases, more data isbecoming available and more reports of injuries caused by deployed
airbags have appeared in the literature (Christy et al., 1998; Jeonand Park, 2006; Audrey et al., 2011; Potula, 2012). For example,McGwin et al. (2003) investigated the associations between a sideairbag and risk of injury in motor vehicle collisions under side
X. Deng et al. / Accident Analysis and Prevention 55 (2013) 232– 241 233
F side ip
ittoaattsata
hvfiefiettwmeubfvgf
(((
((
t
ig. 1. Two kinds of side impact finite element (FE) simulations were used for thearameters on head injury: one with the airbag and one without the airbag.
mpact using the United States national automotive sampling sys-em (NASS) databases for the years 1997–2000; they concludedhat front seat occupants of vehicles with side airbags had a riskf head injury similar to that of occupants of vehicles without sideirbags. Recently, Potula et al. (2012) investigated occupant injuriesnd safety through simulating the interaction between side cur-ain airbag deployment and an out-of-position occupant and foundhat the airbag significantly affects occupant injury/safety. As such,tudies to assess the mechanisms of head injury associated withirbag deployment are being conducted to determine the parame-ers of the airbag and the position of the drivers in relation to theirbag (Kiuchi, 1998; Iyota, 2003; Fang et al., 2005a,b).
Although the aforementioned studies confirm that the airbag-ead separation distance, the airbag deployment, and the airbagariables all have a significant effect on the occupant injury, veryew studies to date have focused on the side curtain airbag andts relationship to head injuries (Ref. Potula, 2012 and Potulat al., 2012). In this study, we seek to obtain the influences ofve parameters on two responses, namely, the peak head accel-ration and the head injury criterion measured over a 36 msime period (HIC36). Because we are considering five parame-ers, the design-of-experiments (DOE) technique (Fisher, 1935a,b)as used to evaluate the effect of each parameter in an efficientanner. Popularized by Taguchi (1987) in the field of quality
ngineering, the DOE approach used herein has been previouslytilized in various contexts of mechanics problems and designy Horstemeyer et al. (1999, 2009). By doing so, the crash per-ormance of the side curtain airbag and the effect of the airbagariables on the occupant injuries were comparatively investi-ated and assessed. The airbag parameters considered were theollowing:
1) The airbag-head separation distance,2) a moving deformable barrier (MDB) strike velocity,3) different trigger times between when the MDB started to strike
the car and the time when the bag started to inflate,4) initial airbag inlet temperature airbag, and5) the number of computational particles employed representing
the amount of air in the airbag.
Hence, the objective of this study was to delineate the rela-ionship between crash severity, head injury of a dummy, and
mpact of a Dodge Neon using a dummy in order to investigate the effect of airbag
airbag-head separation distance for different parameters of a sidecurtain airbag using finite element (FE) simulations.
The outline of this paper is as follows. Section 2 introduces thesimulation models and presents details of the simulation set-up.Section 3 presents a brief overview of the DOE methodology includ-ing the parameters, the parameter values used, and the responsevariables. Section 4 presents the simulation results and the anal-ysis of those results using the DOE methodology. Finally, Section4 presents conclusions and the implications of this study towardsvehicular design of side curtain airbags.
2. Methods
2.1. Side impact simulation
In this section, the full-scale FE model of the Dodge Neon witha side curtain airbag used in the side impact crash simulations isintroduced; the FE model was originally developed and validatedat the National Crash Analysis Center (NCAC) of the United Statesfor a full frontal impact, and the FE model was modified and val-idated for use in simulations of side impacts by the Center ForAdvanced Vehicular System (CAVS), Mississippi State University(Solanki et al., 2004; Fang et al., 2004a,b, 2005a,b; Yildiz and Solanki,2011; Potula, 2012). One kind of side impact meshes with airbagis illustrated in Fig. 1. Fig. 2 illustrates the different airbag param-eters considered in this study. Both side impact simulations wereperformed using a MDB.
As illustrated in Fig. 1, the side-impact simulations wereachieved using an MDB, which impacted the vehicle at differentclosing speeds at an angle of 63◦ with the longitudinal center-line ofthe test vehicle. The wheels of the MDB were “crabbed” 27◦ towardthe rear of the test vehicle to ensure that the front of the MDB wasparallel to the side of the test vehicle at the moment of impact. TheMDB FE model had a total mass of 1388 kg comprised of two hon-eycomb sections, the front face, and the main block. The front faceof the barrier had eight components. Both honeycomb blocks werecovered by shell elements, and null materials were used in defininga contact interface. The previous work on FE model development
includes those of Solanki et al. (2004), Fang et al., 2004a,b, 2005a,b,Yildiz and Solanki (2011), and Potula (2012).In addition, the dummy injury criteria used in the simulationswere the head injury criteria, head acceleration, and the peak
234 X. Deng et al. / Accident Analysis and
Fig. 2. Side impact crash pictures showing a few of the factors used in the fol-lowing analysis: (a) the airbag-head separation distance (factor A), (b) the movingdc
ph3fH
H
˛
whtt
2
iraatac
eformable barrier (MDB) with its strike velocity (factor B), and the number ofomputational particles to represent the amount of air (factor E).
ressure of head experienced from the airbag impinging on theead during the crash. The head injury criterion measured over a6 ms time period (HIC36) were used to assess the influence of dif-erent design variables on the head injury of the dummy model. TheIC36 value is shown in the following equation (www.mdvfs.de):
IC = SUPt1t2
{(1
t2 − t1
∫ t2
t1
a(t)dt
)2.5
(t2 − t1)
}(1)
=√
a2x + a2
y + a2z
here a is the resultant acceleration of the center of gravity of theead in units of acceleration of gravity (g = 9.81 m/s2), and t1 and
2 are the points in time (in seconds) during the crash for whichhe HIC was at a maximum.
.2. Design of experiments methodology
The Taguchi method (Taguchi, 1987) studies the influence of thenput parameters on output responses using a minimal amount ofuns. This approach works whether the objective is a maximization,
minimization, or reaching a particular target. Thereafter, the
nalysis of variance (ANOVA) technique can be used to test the rela-ive influence and statistical significance of the various parametersnd their assigned factor levels. Once the necessary finite elementomputations were performed and the responses were obtained,Prevention 55 (2013) 232– 241
the DOE method was used to screen parameters and obtain thedesired parametric influence (Montgomery and Douglas, 1996).For these reasons, the DOE methodology has been widely usedin crashworthiness optimization studies and engineering analysis(Horstemeyer et al., 2003; Oktem et al., 2007; Chan et al., 2010;Tansel et al., 2011; Yildiz and Solanki, 2011).
The design matrix was generated to sample the main effect of thevarious parameters. Three levels were chosen for the five param-eters, which are denoted by −1, 0, and 1 (minimum, mean, andmaximum, respectively) in Table 1. The various factors and theirlevels are subsequently described. The Taguchi L27 orthogonal arraywas used to sample points in the design matrix. Table 2 shows thevalues of the parameters (columns, A–E) for the 27 runs (rows).The analysis of the responses for the L27 orthogonal array is subse-quently addressed as well.
2.2.1. Airbag-head separation distanceFactor A corresponds to the airbag-head separation distance.
Hallman et al. (2008) observed changes in the viscous crite-rion as great as 135% with changes occupant–airbag distancefrom 0 to 160 mm in steps of 20 mm. Hallman et al. (2010) alsofound that a closer occupant sustained an increased risk of injurycompared to a farther occupant while incrementing the occupant-airbag distance from 20 to 240 mm in steps of 20 mm. In thispaper, the airbag-head boundary separation distance was cho-sen as 5 mm, 65 mm, and 125 mm (−1, 0, 1) when consideringthe estimated dummy position with respect to the side curtainairbag.
2.2.2. Moving deformable barrier (MDB) strike velocityFactor B corresponds to the MDB strike velocity. The MDB
velocity was chosen as 20 mph, 30 mph, and 40 mph (−1, 0, 1)to compare the results of HIC36 and/or peak head accelerationbetween simulation and the federal motor vehicle safety standard(FMVSS) 208. This range of MDB velocity was used, since the testis intended to represent most of the real world crashes (FMVSS208).
2.2.3. Airbag trigger timeFactor C corresponds to the airbag trigger time. The airbag
system components along with one or more inflator stages are trig-gered by accelerometers in combination with sophisticated controllogics. Subject to the type of collision, the trigger times vary from10 ms to 30 ms after the first impact (Gregor and Wolfgang, 2009).In our study, the airbag trigger time is defined as the time betweenwhen the MDB initially strikes the car and the time when the airbagstarts to inflate. The airbag trigger time was chosen as 10 ms, 20 ms,and 30 ms (−1, 0, 1) by considering realistic times of airbag deploy-ment at the time of the impact.
2.2.4. Initial airbag inlet temperatureFactor D corresponds to the initial airbag inlet temperature. The
control volume (CV) method used for continuum formulations inairbag modeling uses the ideal gas law for the constitutive descrip-tion of the airbag gases. The temperature is always adapted for theideal gas law. The particle method in LS-DYNA uses the same data todefine the inflator characteristics as in the CV formulation, i.e. heatcapacities, temperature, and mass flow rates. Internally, the datais automatically converted to particle properties (Potula, 2012). Assuch, the initial airbag inlet temperature was chosen as 260 K, 280 K,and 300 K (−1, 0, 1) based upon realistically-achievable environ-mental temperatures.
The initial condition of the inlet air temperature comes fromthe ambient conditions; however, because the pressure is so highand nitrogen is typically used for the pressure, the temperature willdrop on the order of 10–15◦ over the airbag release in practice. The
X. Deng et al. / Accident Analysis and Prevention 55 (2013) 232– 241 235
Table 1Varying parameters and their levels considered for the finite element simulations.
Factor Process parameter Units Min (level −1) Mean (level 0) Max (level 1)
A Airbag-head boundary separation distance mm 5 65 125B MDB velocity mph 20 30 40C Airbag trigger time ms 10 20 30D Initial airbag inlet temperature K 260 280 300
hit2itdoph
2
erdsPepptcwp
TD
E Number of particles to represent the amount of air
eat transfer is essentially independent of the initial conditions, sof the initial condition is very hot like in a desert in Arizona, theemperature drop will still be about the same. Since one can get22 K in Alaska on a cold day and 322 K in a hot day in Arizona, the
nitial ambient temperature plays a much more dominant role forhe inlet airbag temperature. In this study, we only employed a 40◦
ifference (K) because that covers most of the cases; however, ifne were to consider the two extremes, then the airbag inlet tem-erature influence would even be more than the results indicateere.
.2.5. Number of computational particlesFactor E corresponds to the number of computational particles
mployed with the smooth particle hydrodynamics method thatepresent the air in the airbag. In an airbag, pressure is built up byiscrete particle–fabric impacts. Moreover, particle–particle colli-ions are necessary for a realistic dynamical behavior of the gas.articles are assumed spherical for efficient contact treatment. Forvery 1 l of airbag volume, it is recommended there be 4000–5000articles for predictive results (Olovsson, 2007). The details of the
article method are given by Potula, 2012. Moreover, it is compu-ationally infeasible to model every single molecule inside the sideurtain airbag. In this paper, the number of computational particlesas chosen as 100,000 particles, 150,000 particles, and 200,000articles (−1, 0, 1).able 2esign matrix layout and corresponding results using the L27 orthogonal array (bold font
Simulation number Factors Response(actual)
A B C D E HIC36 Peak h
1 −1 −1 −1 −1 −1 26.9 27.6
2 −1 −1 0 0 0 79.7 52.3
3 −1 −1 1 1 1 311.2 87.9
4 −1 0 −1 −1 0 63.3 29.7
5 −1 0 0 0 1 146.3 65.8
6 −1 0 1 1 −1 484.7 110.0
7 −1 1 −1 0 −1 191.4 48.1
8 −1 1 0 1 0 987.1 101.2
9 −1 1 1 −1 1 362.7 53.2
10 0 −1 −1 1 1 140.3 27.9
11 0 −1 0 −1 −1 83.9 37.2
12 0 −1 1 0 0 83.2 37.2
13 0 0 −1 0 1 149.0 38.7
14 0 0 0 1 −1 180.9 50.5
15 0 0 1 −1 0 211.5 49.9
16 0 1 −1 1 0 1095.0 116.6
17 0 1 0 −1 1 362.7 53.3
18 0 1 1 0 −1 335.5 60.7
19 1 −1 −1 −1 −1 356.4 43.7
20 1 −1 0 0 0 112.6 41.8
21 1 −1 1 1 1 98.6 37.2
22 1 0 −1 1 0 244.30 56.4
23 1 0 0 −1 1 261.2 58.6
24 1 0 1 0 −1 338.2 74.2
25 1 1 −1 0 1 430.7 79.7
26 1 1 0 1 0 457.0 85.3
27 1 1 1 −1 −1 554.6 110.9
100,000 150,000 200,000
2.3. Taguchi analysis
In essence, the Taguchi L27 orthogonal array used here providesa linear system of equations that relate the response{R} from finiteelement computations to the desired influences {ˇ} as
{R} = {P}{b} (2)
where [P] listed in Table 2 is the parameter matrix correspondingto the chosen orthogonal array. The components of {R}, {ˇ}, and{P} are denoted by the following:
{R}
=
⎧⎪⎪⎪⎨⎪⎪⎪⎩
x1
x2
.
.
.
x27
⎫⎪⎪⎪⎬⎪⎪⎪⎭
,{
ˇ}
=
⎧⎪⎪⎪⎨⎪⎪⎪⎩
ˇ0
ˇA
.
.
.
ˇE
⎫⎪⎪⎪⎬⎪⎪⎪⎭
,{
P}
=
⎧⎪⎪⎨⎪⎪⎩
1 −1 −1 −1 −1 −1
1 −1 −1 0 0 0
.
.
....
.
.
....
.
.
....
1 1 1 1 −1 −1
⎫⎪⎪⎬⎪⎪⎭ (3)
The goal is to determine the values for {ˇ}, given that the DOEsetup prescribes [P], and the finite element calculations give theresponse vector {R}. The first column in the parameter matrix[P] given in Eq. (4) relates to ˇ0, a statistical average of the DOEoutputs. The coefficients ˇA, ˇB, ˇC, ˇD,and ˇE reflect the influ-ences of airbag-head separation distance, MDB velocity, trigger
time between when the MDB starts to strike the car and the timewhen the airbag starts to inflate, initial airbag inlet temperature,and the computational particles, respectively, as listed in Table 1.The balance of the Taguchi method is shown in the matrix [P], whererepresents the highest and lowest values).
S/N ratio values for HIC 36 S/N ratio values for peakhead acceleration
ead acceleration
−28.59 −28.82−38.03 −34.37−49.86 −38.88−36.03 −29.46−43.30 −36.36−53.71 −40.83−45.64 −33.64−59.89 −40.10−51.19 −34.52−42.94 −28.91−38.48 −31.41−38.40 −31.41−43.46 −31.75−45.15 −34.07−46.51 −33.96−60.79 −41.33−51.19 −34.53−50.51 −35.66−51.04 −32.81−41.03 −32.42−39.88 −31.41−47.76 −35.03−48.34 −35.36−50.58 −37.41−52.68 −38.03−53.20 −38.62−54.88 −40.90
2 s and Prevention 55 (2013) 232– 241
enbwslroo
ˇ
wi
hoeutssruttra
wtna
3
tmcelsm#dai1bttictgrd
tu#s
Fig. 3. Acceleration versus time calculated at the dummy head from the airbagimpinging on the head during the side impact crash. Three different separationdistances are illustrated for the various cases: (a) with a minimum airbag-head sep-aration distance (simulations #1–#9), (b) with a mean value for the airbag-head
36 X. Deng et al. / Accident Analysi
ach level within any of the remaining columns occur an equalumber of times. This introduces the statistical independence, oralance, into the orthogonal array. If the response {R} associatedith one level changes at another level, then that parameter has a
trong impact on the response being considered. Because differentevels occurs an equal number of times, an effect on the particularesponse of interest by each of the other parameters is canceledut. The approximate values of {ˇ} is achieved using the methodf least squares written as
= (PT P)−1
(PT R) (4)
here the aforementioned coefficients correspond to thosentroduced in Eq. (4) and described previously.
In our study, two sets of responses are considered: the 36 msead injury criterion (HIC36) and the peak head acceleration. Basedn the L27 array, 27 simulations were performed to evaluate theffect of the side curtain airbag on head injuries. Moreover, we alsosed the mean responses for each factor and factor level to converthe simulation results into a value for evaluating the worst-casecenario in the parameter analysis to compare to a car without theide curtain airbag. Finally, we calculated the signal-to-noise (S/N)atios stemming from the design employed. The Taguchi techniquetilizes the S/N ratio to minimize the variances of quality charac-eristics results from the desired value. The smaller the S/N ratio (orhe more negative), the worse the factor and factor level. The S/Natio can be evaluated using the following equation (Montgomerynd Douglas, 1996; Oktem et al., 2007):
S
N= −10 log
(1N
n∑i=1
(yi)2
)(5)
here yi is the value of HIC36 or the peak head acceleration forhe ith test experiment, n is the number of tests, and N is the totalumber of data points. The crashworthiness simulation responsesre analyzed in Section 3.
. Results and discussion
The head acceleration was calculated as a function of time forhe 27 different sets of parameters from the Taguchi L27 design
atrix. An SAE 60 ms filter was used as standard to filter the data forompensate for noise ratio or vibration [www.ls-dyna.com; Christyt al., 1998]. Fig. 4 illustrates the acceleration versus time calcu-ated at the dummy head as the airbag struck the head during theide impact crash. The various subplots correspond to the mini-um, median, and maximum airbag-head separation distances: (a)1–#9, (b) #10–#18, and (c) #19–#27. The differences in Fig. 4emonstrate that there is significant variability present in thecceleration–time histories that the airbag-head impact incurs dur-ng a side impact crash. All peak head accelerations occur between5 and 70 ms from the initial impact. Clearly, the visible differencesetween the various acceleration–time curves may be caused byhe levels of the various other factors, such as the MDB velocity andhe parameters of the side curtain airbag. To facilitate the compar-son between factors in the different curves in Fig. 3, the followingonventions were used. First, the color of the line style correspondso the different factor levels for the MDB velocity (red: 20 mph,reen: 30 mph, black: 40 mph). Second, the different line styles cor-espond to different factor levels for the trigger time (solid: 10 ms,ashed: 20 ms, dotted: 30 ms).
The head acceleration curves shed light on the impact of some of
he factors prior to subsequent analysis. First, there are several sim-lation runs that experience two or three peak accelerations: #1,4, #7, #8, and #25. The second peak has some time delay in theseimulations. These results seem to indicate that the dummy’s riskseparation distance (simulations #10–#18), and (c) with maximum airbag-headseparation distance(simulations #19–#27). Note that the range of accelerationsranged from 30 to 120 G.
to incurring this two/three peak acceleration history is much higherwith a closer airbag-head separation than with further airbag-headseparation (i.e., compare Fig. 3a–c). However, the second peak istypically the more important one for validation, since the first peak
is much lower than the second peak in the simulations #4, #7, #8and #25. These results suggest that there is an interaction betweenthe airbag-head separation distance and the airbag trigger timewith respect to the shape of the acceleration curve.
X. Deng et al. / Accident Analysis and
Fn
soFtFtbouteslutb
ipalmdttftba
rfaaitewlnlpvtc
i
ig. 4. Scatter diagram of HIC 36 and the peak head acceleration versus simulationumber.
Second, notice that in most simulations, the head accelerationtarts to register a response approximately 25 ms after the impactccurs. However, there are a few cases where this is not the case.or instance, and perhaps most notably, the head acceleration startso increase around 15 ms for three simulations (#1, #4, and #7) inig. 3a. The solid lines for these simulations indicate that all havehe minimum airbag trigger time (C), which accelerated the impactetween the side curtain airbag and the dummy’s head. More-ver, the severity of the peak acceleration increases (and the timentil the second peak acceleration decreases) with MDB velocity forhese three cases (i.e., #1–20 mph, #4–30 mph, #7–40 mph). Inter-stingly, compare this case to that of Fig. 3c (furthest airbag-headeparation distance). In this case, the airbag trigger time had veryittle effect on the time to impact. The bigger factor in these sim-lations was the MDB velocity, i.e., the 40 mph velocity impactedhe dummy head much quicker than the 20 mph velocity, as mighte expected.
Third, notice that the MDB velocity plays a commanding rolen the peak acceleration. For Fig. 3b and c, notice that the headeak accelerations are ordered such that the three highest peakccelerations are for an MDB velocity of 40 mph, while the threeowest peak accelerations are for an MDB velocity of 20 mph, as
ight be expected. However, for a short airbag-head separationistance (Fig. 3a), the three highest peak accelerations differ forhree different MDB velocities. Clearly, the complex acceleration-ime histories are governed by interactions between the variousactors and factor levels as well as the main effects. For instance,he aforementioned examples have all shown how interactionsetween airbag-head separation distance (A), MDB velocity (B), andirbag trigger time (C) affect these histories.
Consequently, these complex acceleration-time histories wereeduced to facilitate analysis of the effect of the various factors andactor levels. The head injury criteria, HIC36, and the peak headccelerations, along with their S/N ratios, were calculated from thecceleration–time curves (Fig. 3). Fig. 4 shows the two correspond-ng values for both HIC36 and the peak head accelerations for thewenty-seven simulations. As shown in Fig. 3, five peak head accel-rations (#3, #6, #8, #16, #27) and two HIC36 values (#8 and #16)ere found to be much higher than the remainder of the simu-
ations. This plot shows that high peak head accelerations do notecessarily mean high HIC36 values, although the two are corre-
ated, to some extent. Table 2 summarizes the DOE results for theeak head accelerations and the HIC36, which are the responseariables for the present DOE design. As shown in Figs. 3 and 4,
he highest peak head acceleration was observed at the 16th testondition in Table 2.The crashworthiness simulations were also visualized to exam-ne the differences between the different factors. For example, Fig. 5
Prevention 55 (2013) 232– 241 237
illustrates the sequential displacement and deformation history at10 ms, 40 ms, 70 ms, 110 ms, and 150 ms for simulations #1 and#16. Table 2 lists the factor levels for the two cases with the low-est and highest values of HIC36 and the peak head acceleration.For simulation #1, the dummy’s position does not shift very muchat all in comparison to simulation #16. While the factor levelshave certainly influenced this drastic change in response betweenthe two cases (simulation #16 has a higher MDB velocity, higherairbag-head separation distance, higher initial temperature insideairbag, and higher number of particles, but a lower trigger time),it is difficult to identify exactly which factor or factor level causedthe difference without the subsequent analysis. Since the HIC36and peak head acceleration appear to be adequate measures of thetrauma to the head (Fig. 5), the remainder of this paper will con-centrate on analyzing how the factors and factor levels contributeto the severity of the side impact to the dummy.
3.1. ANOVA analysis
The relative influences of the various factors can be quanti-fied using Eq. (5). The parameters{ˇ} for the HIC36 and the peakhead acceleration are listed in Table 3. Based upon the Table 3,Fig. 6 shows the normalized parametric influences on the HIC36and the peak head acceleration, respectively, from the DOE analysisretrieved from the head injury criterion and the peak head accelera-tion data listed in Table 2. The MDB velocity was the most influentialparameter in both cases, followed by the initial temperature insidethe airbag. Interestingly, the initial airbag temperature was only∼31% lower of an influence than the MDB velocity and the triggertime was only ∼54% lower of an influence than the MDB velocitywhen considering the peak head accelerations. When consideringthe wide range in MDB velocities used in this study, a designer usethis information to help mitigate damage caused in a side impactcrash by higher MDB velocities. Moreover, most factors contributesimilarly to both the HIC36 criterion and the peak head accelera-tion. However, the airbag trigger time’s contribution to the HIC36criterion is much lower compared to its influence on the peak headacceleration.
Another method for analyzing the data in Table 2 is to employanalysis of variance (ANOVA) techniques. The ANOVA analysis wasdone and is presented for HIC36 and peak head acceleration andthe results are shown in Tables 4 and 5, respectively. Here, theANOVA analysis concentrates on the main effects plus any two-factor interactions that were found to be significant (based on 95%confidence limit). A typical method for interpreting an ANOVA anal-ysis is to look at the percent contribution of each factor (based onthe percent of the total sum of squares attributed to each factor). Forboth the HIC 36 and the peak head acceleration, the MDB velocitycontributed the most to the variance in results (43.2% and 30.8%,respectively). Interestingly, the airbag inlet temperature was thesecond highest main effect contributor at 10.5% and 12.9%, respec-tively. Additionally, there are a few two-factor interactions thatwere significant. For example, the interaction between airbag-headseparation distance and airbag inlet temperature (A*D) as well asthe interaction between MDB velocity and airbag inlet tempera-ture (B*D) contributed to 16.6% and 12.2% of the variance in HIC36(Table 4). Additionally, the interaction between airbag-head sep-aration distance and airbag inlet temperature (A*D) contributedto 21.0% of the variance in peak head acceleration (Table 5). Theaforementioned main effects and two-factor interactions are allstatistically significant based on a 95% confidence limit. Finally,the ANOVA tables also show that several factors had only minor
contributions to HIC36 and peak head acceleration: airbag-headseparation distance (A), airbag trigger time (C), and number of parti-cles (E). While these factors were not statistically significant, this isdue, in part, to the strength of the other factors and may change with
238 X. Deng et al. / Accident Analysis and Prevention 55 (2013) 232– 241
Fig. 5. Frontal view of side impact simulation results of the dummy-interactive model with the side curtain airbag at (a) 10 ms, (b) 40 ms, (c) 70 ms, (d) 100 ms, (e) 150 ms.Simulation #1 and #16 correspond to the lowest and highest HIC36 and peak head acceleration values, respectively. Note the dummies positional changes during the timingsequence under the two different circumstances.
Table 3Peak head and HIC acceleration responses from the finite element simulations for the coefficients in Eq. (5).
Coefficient HIC36 Peak head acceleration
ˇ0 ˇA ˇB ˇC ˇD ˇE ˇ0 ˇA ˇB ˇC ˇD ˇE
Values 301.81 11.13 193.55 3.11 98.31 −26.85 60.58 0.67 17.57 8.25 12.08 −4.25
X. Deng et al. / Accident Analysis and Prevention 55 (2013) 232– 241 239
Table 4Analysis of variance for HIC36.
Factor Sum of squares % contribution Degrees of freedom Mean square F ratio p-Value
A 3149 0.2 2 1574.5 0.17 0.8495B 741,915.8 43.2 2 370,957.9 39.22 0.0001a
C 66,578.6 3.9 2 33,289.3 3.52 0.0801D 180,640.5 10.5 2 90,320.3 9.55 0.0076a
E 27,413.8 1.6 2 13,706.9 1.45 0.2903A*D 285,917.7 16.6 4 71,479.4 7.56 0.0080a
B*D 208969.2 12.2 4 52242.3 5.52 0.0197a
Error 75662.2 4.4 8 9457.8Total 1717470.5 100 26
a Statistically significant based on 95% confidence limit (p-value < 0.05).
Table 5Analysis of variance for peak head acceleration.
Factor Sum of squares % contribution Degrees of freedom Mean square F ratio p-Value
A 901 5.0 2 450.52 1.58 0.2466B 5576.2 30.8 2 2788.12 9.76 0.0030a
C 692.3 3.8 2 346.15 1.21 0.3317D 2345.1 12.9 2 1172.53 4.1 0.0439a
E 485.9 2.7 2 242.95 0.85 0.4515A*D 3804.3 21.0 4 951.07 3.33 0.0472a
Error 3429 18.9 12 285.75
dst
ifHata
FabstiMf
Total 18,132.8 100 26
a Statistically significant based on 95% confidence limit (p-value < 0.05)
ifferent factor levels (e.g., this is not an indication that airbag-headeparation distance has no effect, just little effect when comparedo the other factors).
Fig. 7 (HIC36) and Fig. 8 (peak head acceleration) show thendividual data points and mean response curves for the differentactors and factor levels in the L27 design matrix. In Fig. 7, the meanIC36 value increases with increasing MDB velocity (B) and initial
irbag inlet temperature (D); the two highest HIC values correspondo high factor levels in both factors. However, the airbag-head sep-ration distance (A), the airbag trigger time (C), and the number ofig. 6. Normalized design of experiments (DOE) results for HIC36 and the peak headccelerations: (A) the airbag-head separation distance, (B) the moving deformablearrier (MDB) strike velocity, (C) trigger time between when the MDB started totrike the car and the time when the bag started to inflate, (D) initial airbag inletemperature to the airbag, and (E) the number of computational particles represent-ng the amount of air in the airbag. Note that the primary influence factor was the
DB strike velocity followed by the initial airbag inlet temperature to the airbagollowed by the trigger time.
particles (E) have very little effect based on the mean response. InFig. 8, the mean peak acceleration increases with increasing MDBvelocity (B), airbag trigger time (C), and initial airbag inlet tem-perature (D). Interestingly, referring to Table 5, though, the airbagtrigger time was not statistically significant; this most likely resultsfrom the high peak head acceleration corresponding to the low levelfor the airbag trigger time. Aside from this factor, the responses inFigs. 7 and 8 correlated well with the statistically significant factorsfrom the ANOVA analysis (Table 4 and Table 5).
An indicator of the crashworthiness of a vehicle is often asso-ciated with a HIC36 and/or peak head acceleration values for thehead of a dummy model (based upon FMVSS 208). The lower the
HIC36 and/or peak head acceleration, the better the crashworthi-ness of the vehicle. Table 6 lists the mean responses shown inFigs. 7 and 8. From the results in Table 6, the best design parame-ters for HIC 36 are A0 (airbag-head separation distance = 65 mm),Fig. 7. Mean response graph-variation of HIC36 with different process parameters.The MDB velocity (B in this case) has the largest influence on the mean response.
240 X. Deng et al. / Accident Analysis and Prevention 55 (2013) 232– 241
Table 6Response mean values table (Bold font represents the highest values).
Factor symbol HIC36 The peak head acceleration
A B C D E A B C D E
Level −1 294.81 143.65 299.70 253.69 283.61 63.98 43.64 52.04 51.57 62.541
5
BDcaatashiMsfwmmtt
3
atawlHdt(
Fpt
Level 0 293.56 231.04 296.83 207.4Level 1 317.07 530.74 308.91 444.3Mean value for all 301.81
−1 (MDB velocity = 20 mph), C0 (airbag trigger time = 20 ms),0 (initial airbag inlet temperature = 280 K) and E1 (number ofomputational particles = 200,000). Moreover, for the peak headcceleration, the best design parameters are A0 (airbag-head sep-ration distance = 65 mm), B−1(MDB velocity = 20 mph), C−1 (therigger time = 10 ms), D−1 (initial airbag inlet temperature = 260 K)nd E1 (number of computational particles = 200,000). There areeveral factor levels that are consistent for both HIC36 and peakead acceleration: airbag-head separation distance (A), MDB veloc-
ty (B), and number of computational particles used (E). The lowerDB velocity is certainly not surprising. However, the airbag-head
eparation distance suggests that certain out-of-position locationsor the dummy may reduce the chance of head trauma, agrees wellith the results of Potula et al. (2012). The number of particleserely represents some of the variability in the airbag deploymentethod. Also, notice that the best factor level for the airbag trigger
ime (C) and initial airbag inlet temperature (D) are different for thewo different measures.
.2. Determination of the worst-case scenario with airbag
The DOE study indicated that the MDB velocity, the initialirbag inlet temperature, and the airbag trigger time, and a fewwo-factor interactions were influence parameters for head injuryt the time of the crash. Table 6 lists the mean response values,hich will be used in the subsequent analysis to find the worst
evels for five parameters. The worst design parameters for both
IC and peak head acceleration are A1 (airbag-head separationistance = 125 mm), B1 (MDB velocity = 40 mph), C1 (the triggerime = 30 ms), D1 (initial airbag inlet temperature = 300 K) and E0number of computational particles = 150,000). Interestingly, theig. 8. Mean response graph-variation of the peak head acceleration with differentrocess parameters. The MDB velocity (B in this case) has the largest influence onhe mean response.
370.42 52.44 59.31 60.67 55.39 63.37251.41 65.31 78.78 69.02 74.78 55.81
60.58
highest mean responses are the same for both HIC and peak headacceleration. Consequently, the worst-case predictions can becalculated using the mean responses for the different factors andfactor levels (Table 6) along with the following procedure (Oktemet al., 2007; Tansel et al., 2011).
Predicted maximum peak head acceleration= A1 + B1 + C1 + D1 + E0 − 4 × (R̄)= 65.31 + 78.78 + 69.02 + 74.78 + 63.37–4 × (60.58) = 145.57
Predicted maximum HIC36= A1 + B1 + C1 + D1 + E0 − 4 × (R̄)= 317.07 + 530.74 + 308.91 + 444.35 + 370.42–4 × 301.81 = 764.25
where R̄ is the mean response value for HIC36 and the peak headacceleration. Based on the FMVSS 208 (amended by the NationalHighway Traffic Safety Administration), the peak head accelerationmust not exceed 80 G and the HIC36 must not exceed 1000 at anMDB velocity of 40 mph. The above analysis shows that it is verydangerous for this vehicle with the worst-case parameters for theside curtain airbag, based on the peak head acceleration values.Therefore, the initial airbag inlet temperature, the trigger time, andthe airbag pressure should be incorporated into vehicular designprocess when optimizing for the head injury criteria in the furtherstudy.
4. Conclusions
In this paper, the head injury of the hybrid III 50th percentiledummy in side impact crash scenarios was investigated usingvarious factors within a design-of-experiment Taguchi approach(Table 2). Large-scale finite element simulations were conducted toelucidate how various parameters affected two head injury criteria:the HIC36 value and peak head acceleration. Finally, the worst casescenario for the various parameters was predicted to quantify theinfluence of the side curtain airbag. In particular, the main findingsof this study are the following:
(1) The airbag-head separation distance, airbag trigger time, andMDB velocity all affected the head acceleration versus timecurves (Fig. 3). For instance, simulations indicated that the min-imum airbag trigger time accelerated the impact between theside curtain airbag and the dummy’s head for low MDB veloc-ities but not necessarily for higher MDB velocities. Moreover,the MDB velocity correlated well with the peak head accelera-tion for longer airbag-head separation distances, A = {0, 1}, butnot necessarily for the lowest airbag-head separation distance.Finally, a closer airbag-head separation increased the dummy’srisk to incurring a two/three peak acceleration history. Theseobservations all indicate that both the main effects and interac-tions between the various factors examined herein have a stronginfluence on the risk of head injury incurred by the occupant.
(2) The ANOVA results (Tables 4 and 5) show that the most influ-ential factors for HIC 36 and peak head acceleration are theMDB velocity (B) followed by the initial temperature insidethe airbag (D) and the trigger time. Additionally, the combined
s and
(
stoi
A
VttKvCE0
A
t
R
A
C
CD
F
X. Deng et al. / Accident Analysi
(A*D) airbag-head separation distance (A) and initial airbagtemperature (D) was a statistically significant interaction forboth criteria. Finally, the interaction (B*D) between the MDBvelocity (B) and airbag inlet temperature (D) was a statisticallysignificant interaction for HIC36. Based on the sum of squares,the percent contributions for the HIC36 are the following: B(43.2%), D (10.5%), A*D (16.6%), B*D (12.2%). Likewise, the per-cent contributions for the peak acceleration are the following: B(30.8%), D (12.9%), A*D (21.0%). The remaining factors only hadminor effects related to the variation of the simulation results.
3) The worst-case scenario for the side curtain airbag waspredicted using the DOE matrix results and validated byan additional crashworthiness simulation with the extremeworst case parameters. The DOE predicted peak accelerationwas142 G only 3% different than the additional FE simulationresult of 146 G, while the DOE predicted HIC36 value was 764,the additional FE simulationwas638, approximately 20% lowerthan the predicted value.
In light of the results presented here, future investigationshould incorporate vehicle-specific airbag designs that optimizehe airbag parameters for injury mitigation. Other out-of-positionccupant orientations may also be relevant but were not performedn this study [45].
cknowledgments
The authors would like to acknowledge the Center for Advancedehicular System (CAVS) at Mississippi University for support of
his work. The authors would also like to thank the support ofhe Spring Sunshine Plan for the grant (grant no. 10202258), theey Disciplinary of Mechanical Design and Theory of Xihua Uni-ersity (XZD0813-09-1), Talent Introduction of Xihua University,hina (Grant No. Z1220217) and the Department of Energy Nationalnergy Technology Laboratory under Award Numbers DE-FC26-2OR22910.
ppendix A. Supplementary data
Supplementary data associated with this article can be found, inhe online version, at http://dx.doi.org/10.1016/j.aap.2013.03.016.
eferences
udrey, H.D., Federic, R., Pacaux, M.P., Herve, M., 2011. Determination of pre-impactoccupant postures and analysis of consequences on injury outcome. Part: Adriving simulator study. Accident Analysis and Prevention 43 (1), 66–74.
hristy, M., Margaret, H., Hines, A.L., Roger, A.S., 1998. Assessment of forearminjury due to a deploying driver-side airbag. In: Proceeding of 16th Interna-tional Technical Conference on the Enhanced Safety of Vehicles, no.98-S5-O-09,pp. 1044–1054.
rash Analysis Criteria description. www.mdvfs.deix, J., Sagawa, K., Sahare, L., Hanmound, S., Fulk, D., Cardinali, A., 2011. A study of
occupant ejection mitigation in side impact crashes. SAE International Journal
of Passenger Cars – Mechanical Systems 4 (1), 91–99.ang, H., Solanki, K.N., Horstemeyer, M.F., Rais-Rohani, M., 2004a. Multi-impactcrashworthiness optimization with full-scale finite element simulations. In:Computational Mechanics: The 6th World Congress on Computational Mechan-ics in Conjunction APCOM.
Prevention 55 (2013) 232– 241 241
Fang, H., Solanki, K.N., Horstemeyer, M.F., 2004b. Energy-based crashworthi-ness optimization for multiple vehicle impacts. In: Proceeding of InternationalMechanical Engineering Congress and Exposition, pp. 2004–59123.
Fang, H., Solanki, K., Horstemeyer, M.F., 2005a. Numerical simulations of multiplevehicle crashes and multidisciplinary crashworthiness optimization. Interna-tional Journal of Crashworthiness 10 (2), 161–171.
Fang, H., Rais-Rohani, M., Liu, Z., Horstemeyer, M.F., 2005b. A comparative studyof metamodeling, methods for multiobjective crashworthiness optimization.Computers and Structures 83 (25/26), 2121–2136.
Fisher, R.A., 1935a. Statistical methods for research workers. Oliver and Boyd.Fisher, R.A., 1935b. The design of experiments. Oliver and Boyd.Fujiyama, Y.J., Sonoyama, D., Obayashi, K., Yu, Q., 2011. Application of CAP to ana-
lyze mechanisms producing dummy injury readings under US side impact testconditions. SAE 2011 World Congress and Exhibition, Detroit, MI, USA, (PaperNumber.2011-01-0014).
Gabauer, D.J., Gabler, H.C., 2010. The effects of airbags and seatbelts on occu-pant injury in longitudinal barrier crashes. Journal of Safety Research 41 (1.),9–15.
Gregor, G., Wolfgang, S., 2009. Improvement of airbag performance through pre-triggering. ESV 2009. (Paper No.09-0229).
Hallman, J.J., Yoganandan, N., Pintar, F.A., 2008. Torso side airbag out-of-positionevaluation using stationary and dynamic occupants. Biomed SciInstrum 44 (6),123–128.
Hallman, J.J., Yoganandan, N., Pintar, F.A., 2010. Toracic injury metrics withside airbag: stationary and dynamic occupants. Traffic and Prevention 11 (4),433–442.
Horstemeyer, M.F., McDowell, D.L., McGinty, R., 1999. Design of experimentsfor constitutive model selection: application to polycrystal elastoviscoplas-ticity. Modelling and Simulation in Materials Science and Engineering 7 (2),253–273.
Horstemeyer, M.F., Ramaswamy, S., Negrete, M., 2003. Using a micromechan-ical finite element parametric study to motive a phenomenological macroscalemodel for void/crack nucleation in aluminum with a hard second phase.Mechanics of Materials 35 (7), 675–687.
Iyota, T., 2003. The effect of occupant protection by controlling airbag and seat-belt. In: 18th International Technical Conference on Enhanced Safety of Vehicles(ESV), vol. 4, issue 3, pp. 2518–2525.
Jeon, S.K., Park, G.J., 2006. Injury patterns of the hybrid III fifth percentile femaledummy according to airbag inflation. Proceedings of the Institution of Mechan-ical Engineers, Part D: Journal of automobile engineering 220 (1), 15–25.
Kiuchi, T., 1998. A simulation study on inflation induced injury and NCAP withdepowered airbag. In: 16th International Technical Conference on EnhancedSafety of Vehicles (ESV), vol. 3, no. 5, pp. 2518–2525.
McGwin, G., Metzger, J., Porterfield, J., Moran, S., Rue, L., 2003. Association betweenside airbags and risk of injury in motor vehicle collisions with near-side impact.Journal of Trauman 55 (3), 430–436.
Montgomery, Douglas, C., 1996. Design and analysis of experiments. ISBN: 0-471r-r15746-5.
National Highway Traffic Safety Administrations (NHTSA). http://www.nhtsa.gov/FARS
Oktem, H., Erzurumlu, T., Uzman, I., 2007. Application of Taguchi optimizationtechnique in determining plastic injection molding process parameters for athin-shell part. Materials and Design 28 (4), 1721–1728.
Olovsson, L., 2007. Corpuscular method for airbag deployment simulation in LS-DYNA. Technical Report R32S-1. IMPETUS.
Potula, S.R., 2012. Safety counter measures: a comprehensive crashworthiness studyof out-of-position (OOP) airbag deployment and passenger impact, Master the-sis. Mississippi State University.
Potula, S., Solanki, K.N., Oglesby, D.L., Tschopp, M.A., Bhattia, M.A., 2012. Investigat-ing occupant injuries and safety through simulating the interaction between sidecurtain airbag deployment and an out-of-position occupant. Accident Analysisand Prevention 49, 392–403, http://dx.doi.org/10.1016/j.aap.2012.03.007.
Solanki, K.N., Oglesby, D.L., Burton, C.L., Fang, H., Horstemeyer, M.F., 2004. Crashwor-thiness simulations comparing PAM-CRASH and LS-DYNA. The 2004 SAE WorldCongress, Dearborn, MI, pp. 8–35, ISBN 094124301X.
Taguchi, G., 1987. System of Experimental Design, vol. IandII. UNIPUB, New York.Tansel, I.N., Gulmez, S., Demetgul, M., Aykut, S., 2011. Taguchi method-
GONNS integration: complete procedure covering from experimental designto complex optimization. Journal of Expert Systems with Application 38 (5),4780–4789.
Yildiz, A., Solanki, K.N., 2011. Particle swarm algorithm based approach for multi-objective crashworthiness and manufacturing optimization. IJAMT 1–10, 201.
