Академический Документы
Профессиональный Документы
Культура Документы
Materials
& Design
Materials and Design xxx (2005) xxx–xxx
www.elsevier.com/locate/matdes
Abstract
This paper is devoted to study and verify the suitability of the expanded polypropylene (EPP) foam as a liner for motorcycle
helmet and to perform helmet design optimization. This EPP foam has a multi-impact protection performance and also has a poten-
tial for ventilation system improvement due to its resiliency. This resiliency allows for the ease of ventilation holes and channels
molding without the foam breakage at the stage of mold extraction. The large scale, non-linear, dynamic finite element package
LS-DYNA3D is used as a verification tool for motorcycle helmet design. Then the simulation work is carried further to provide
data for helmet design analysis and optimization using the response surface methodology (RSM). The foam thickness, the foam
density, and shell thickness being selected as the design factors for the response surface generation and design optimization. The
results showed that the EPP satisfies the 300g headform center of gravity acceleration limit required by most of the international
standards. The extended simulation output data is then used to create the response surface and determine the optimum design
points. Therefore, two main contributions on motorcycle helmet design are achieved. The first is on the use of the EPP as a helmet
energy absorption liner for the motorcycle helmet application, and the second is that by combining the simulation output with the
design of experiment (DOE) method to study the effects of the various factors on helmet design optimization.
2005 Elsevier Ltd. All rights reserved.
*
Corresponding author. Tel.: +603 89466330; fax: +603 86566061. The expanded polypropylene (EPP) foam shows
E-mail address: hamouda@eng.upm.edu.my (A.M.S. Hamouda). strong potential in overcoming such problems
0261-3069/$ - see front matter 2005 Elsevier Ltd. All rights reserved.
doi:10.1016/j.matdes.2005.04.015
ARTICLE IN PRESS
[9,20,12]. However its suitability as a motorcycle helmet namic simulation, which will have a more realistic
liner needs to be thoroughly investigated from the meaning for motorcycle helmet crash studies. Yetham
mechanical performance point of view. Another objec- et al. [31] have developed a simplified finite element
tive which is included in this work is the helmet design model for studying motorcycle helmets using LS-DY-
optimization. This later objective was a challenging task NA3D code. They model the impact of a helmet fitted
in the previous research work due to the complex nature on a wooden headform. The headform was given a
of the helmet structure and the various factors involved. simplified spherical shape, while the helmet was given
Therefore, the two main objectives, which are tar- a hemispherical shape. Various types of helmet shell
geted in this paper, namely; the verification of the materials were used in the analysis, such as GRP,
EPP foam as a suitable liner for motorcycle helmet PC, and LDPE. However, the widely used ABS shell
application, while the later is to develop helmet design material was not considered. Three densities of EPS
optimization. The finite element simulation was the foam were used. Their study has provided an excellent
most suitable choice for such objectives, as once the start of motorcycle helmet simulation and parametric
design is found to be acceptable; the model parame- studies using the finite element simulation techniques.
ters can easily be changed for parametric study. This Recently, Liu et al. [20] carried more advanced helmet
is not feasible experimentally even if few EPP foam crash simulation using the same LS-DYNA3D code.
helmet prototypes where manufactured. This is be- They have used ABS material for the shell and the
cause that any required change in the helmet design EPS foam as the helmet liner. Their main concern
parameters needs some sort of modification to the was to develop a helmeted-head finite element model
molding facility, which is generally not feasible or ex- and carry out a validation process rather than perform-
tremely cost offensive. ing parametric studies or optimization. The mechanical
In the next paragraphs a brief review about the previ- properties for these components were obtained experi-
ous studies on the motorcycle helmet design and analysis mentally under quasi-static conditions. Their simula-
is presented. tion was carried out using both standard headform
At the early stage, solution of the helmeted-head im- model and biomechanical head model. The impact site
pact problem was tackled either by experimental test- was rotated between the front to the back of the hel-
ing or by using analytical methods. The later met. The heights of impact simulated were relatively
techniques involves the development of simplified ana- small, and not conforming to any well known helmet
lytical models, which are composed of springs and testing standards. However, their study is another
dashpots to represent the helmet components. Then essential step in this field. In our study the previously
the analytical model is solved by basic dynamics meth- described modeling techniques can be considered in
ods. In the experimental investigation the results were the development of the EPP foam liner helmet model.
restricted to varying the impact parameters such as The material model will be different from those models
the impact speed or the shape of the anvil. Varying hel- utilized for the EPS.
met parameters experimentally were impossible due to In this study the finite element method is used for the
testing sample manufacturing constraints. This had helmet impact solution. The choice of using the finite
limited the benefits gained from the experimental stud- element method is taken because of its capability to
ies to the standards requirements satisfaction. In the analysis metal and non-metal structures in an accurate
analytical solution, helmet impact on the crown against way, and its availability as a finite element package
flat anvil was analytically modeled by Gilchrist and (such as LS-DYNA3D), coupled with easy access to
Mills [14]. The solution obtained, which is usually the required computing resources.
either one or two dimensional, had limited advantages The design analysis and optimization is performed
due to the approximation involved and the disability of using the response surface method (RSM) in the de-
representing most of the essential impact features sign of experiment (DOE) statistical methods. This
encountered in real accidents. method is easy to apply and provides a comprehensive
The problem of helmet impact modeling and crash overview on the factors effecting a targeted response
simulation is facilitated after the tremendous develop- [27]. Also it is the only method to investigate interac-
ment in the computer technology and the development tion effects between the considered factors [28]. This
of the advanced simulation packages such as the LS- method is extensively used for experimental work,
DYNA3D [21] and the PAM CRASH [7]. and its use for simulation oriented research work is
Vetter et al. [30] conducted a finite element simula- a recent trend. In this regard, few programs are re-
tion of football helmet. Their study involves the simu- cently developed that uses the RSM with the finite ele-
lation of helmet subjected to a static load on the ment simulation as a predicted response. Among those
crown. The effects of the variation of helmet structural are the LS-DYNA-OPT [21]. Therefore it is selected in
and material properties were studied. Some principles this research work to perform helmet design analysis
of their research methodology can be utilized for dy- and optimization.
ARTICLE IN PRESS
2. The research methodology to the response surface method requirements. The peak
linear acceleration, which is the predicted response in
The methodology followed in this study is shown in the DOE, is recorded in each case. The simulation data
the flowchart of Fig. 1. First, in order to model the hel- are inserted in the response surface design matrix, then
met components, the material properties for each com- the response surface is created. After the response sur-
ponent needs to collected. Then helmet model can be face creation, the surface can be analyzed, and the 3 D
built using Hypermesh pre-processor. After model response surface and contours can be plotted. Also opti-
translation to LS-DYNA3D, the helmet model impact mization process can be performed. In this research
is simulated using LS-DYNA3D finite element solver; work both tasks are performed to have a comprehensive
the output is displayed using both LS-DYNA3D Finite helmet design analysis and optimization. Results are
Element Model Builder (FEMB) for the basic results then discussed, and conclusions are withdrawn.
visualization and the Hypermesh post-processor [1] for
more detailed display features. The simulation results
are categorized to two types. The first is the helmet de- 3. Material properties
sign verification simulations while the second is the re-
sponse surface DOE simulations. In the former the Modeling of the complete helmet requires the input
model is checked against the Malaysian helmet standard data of the constituentÕs materials of the helmet. As
MS 1: 1996 which quotes the 300g peak acceleration lim- mentioned previously, the two main components of
it of the headform center of gravity. However, this limit the helmet are the shell and the liner. As the target in
is the same for other international helmet standards this study is the energy absorption of the new foam can-
such as the British standards BS 6658: 1985 [3]. In the didate, the shell material is selected to be the current
later simulations, the impact anvil is kept to the flat type market dominant shell material which is the ABS. This
and the helmet design parameters are varied according material has good impact properties, and possesses a
cost effective manufacturing method. It has some draw
back which is discussed elsewhere under the same re-
Start
search program [9,10]. The use of this material allows
Data Collection
for comparing results with helmets made from the same
shell material and the current EPS foam liner with more
Model Building accurate judgment than composite shell. The liner is the
Pre-processing material under investigation which is the EPP foam. The
( Hypermesh program ) EPP foam can be manufactured with the bead molding
process as for the EPS. Therefore, the most crucial liner
Processing design constraint which is the manufacturability is elim-
Standards Simulations
verfication required by inated from the comparison, and more realistic judg-
simulation the RSM ment can be made on the mechanical performance.
40
X f ¼ X s Un ; ð1Þ
35
30 where U is the relative foam property with respect the
Stress MPa
25 solid-state property.
20 Farnnel ‘‘Xs’’ refers to the property of the solid matrix param-
15 Helmet
eter and the exponent ‘‘n’’ depends on the property in
10 Dean&Read
equation. It is usually lies in the range of 1.0 < n < 2.0
5
0
and for the closed cell EPP foam it takes the following
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 values: for strength = 2, for modulus = 2, and for the
Strain % shear stress = 1 [13].
Fig. 2. ABS experimental tensile stress–strain curves. The first is for a
The non-linear elasticity region: Linear elasticity is
flat plate cut from ABS sheet produced by FARNNEL Company, the limited to small strains, typically 5% or less. The EPP
second is specimens cut from SolidGold helmet shell, while the third is foam can be compressed much larger than this. The
the literature data [6]. deformation is still recoverable (and thus elastic), but
it is non-linear. In compression the stress strain curve
other published test results [6] are very close particularly shows an extensive plateau stress starting form O0el the
in the yield stress, and the youngÕs modulus. Therefore elastic collapse stress. This elastic collapse is caused by
the general trend of later data are considered for the re- buckling of cell walls. For the closed cell EPP foam
gion beyond the tested range. the compression of the gas within the cells, together with
the membrane stresses, which appears in the cell faces,
3.2. Mechanical properties of EPP foam gives a curve of stress which rises with the strain.
At present a general model, which predicts all the ef-
The quasi-static mechanical properties of the EPP fects for the EPP mechanical behavior does not exist.
foams are reasonably available in recent publications, This may be due to the complex nature of the foam
whereas impact properties are very limited [13]. This behavior and the large number of factors, which affect
might be attributed to the fact that low density EPP its behavior. Factors such as density, temperature, strain
foam type is newly developed and not much literature rate, and manufacturing process are found to have a
is published on its characteristics. As the required data greater effect than others [13,14].
of this type of foam is not sufficiently available in the lit- A theoretical expression for the yield stress increase
erature, more detailed investigation and analysis is nec- based on cell gas compression was initially developed
essary. Accordingly the experimental data quoted by by Rusch [11] for closed cell foams and then verified
various references is described in Appendix A, and foam by several researchers ([13,26]).
mathematical modeling theoretical review is also in- This expression uses the isothermal compression of
cluded in this work. the air in the closed cells. The gas in the cells obeys Boy-
Theoretical modeling: The mathematical approaches leÕs law, initially its at the atmospheric pressure of
to model the static and dynamic behavior of the closed p0 = 0.1 MPa (1 atm) and it occupies a fractional vol-
cell semi-rigid EPP foam are described. Typically, the ume (1-D) of the foam. When the compressive strain is
EPP foam stress–strain behavior can be divided into e the gas in the foam exerts a pressure p on the loading
three regions. The first region is the linear elasticity, surface equals to:
the second is the non-linear or long plateau elasticity,
p0 e
and the third is the densification. Mathematical model- p¼ . ð2Þ
ing of the first two regions is the most essential require- 1eD
ment for the foam finite element modeling. Therefore it If there is also a constant contribution of r0 to the
will be descried in more details hereafter. yield stress due to the compressive buckling of the cell
The elasticity region: Theoretical modeling of the elements, the compressive stress is given by
highly non-linear polymeric foams is still a wide area p0 e
for further studies. Most of the previous work was car- r ¼ r0 þ . ð3Þ
1eD
ried out for the static behavior of foams. Many empiri-
cal and semi-empirical equations have been developed, The gas pressure contribution explains part of the
and some of which are applicable for certain foams stress increase at higher strains. Mills [24] reported that
and under specific conditions [13]. by plotting the compressive stress for polyfin foams such
Eq. (1) is a simple and widely used relation which pre- as expanded polypropylene, or polyethylene against the
dicts the elastic mechanical properties of EPP foam. strain function (the right part excluding r0) in Eq. (3)
This equation correlates the mechanical properties of above, the slope p values ranges from 0.08 to 0.2 MPa.
the foam to the density. For mechanical properties Xf The high pressure values can be explained by saying that
(strength, modulus, etc.) a power law relationship has other processes, such as induced cell face tensions, are
been found to apply in most cases equivalent to the foam being filled by a gas at a pressure
ARTICLE IN PRESS
greater than atmospheric. It means that the parameter p ments and robust algorithms for adaptively controlling
in Eq. (3) must be treated as a curve fitting constant, and the solution process. The transient analysis is performed
therefore its necessary to find its value as a function of using an explicit direct time integration procedure and
the foam density for each range of foams. thereby avoids the need for matrix evaluation, assembly
Recently, Mills and Gilchrist modeled the impact of and decomposition at each time step as required by
the BASF EPP foam with Eq. (3) [26]. The foam density many implicit time-integration algorithms. LS-DYNA
range used is 20–90 kg/m3, and the gas pressure is automatically examines the finite element mesh and
0.2 MPa. There results from the mathematical model material properties in order to determine an appropriate
show good correlation with the experimental results. time step size for numerical stability. This time step size
Therefore, the same approach is used in this research is then automatically adjusted throughout the transient
and stress–strain curves for EPP foam are generated analysis to account for contact local material and geo-
using Eq. (3) and the results are shown in Fig. A.1(a) metric non-linearities.
of Appendix A. As can be noticed the theoretical curves
reasonably well approximates the BASF experimental 4.1. Finite element modeling of ABS shell material
results of Fig. A.1(b) and also very close to reported
data by Mills [24] which is shown in Fig. A.2 of Appen- According to test results described previously for the
dix A. These results are also found to well agree with ABS shell material, the stress–strain data are averaged
data published by other researchers for the same type taking into consideration the test results of the helmet,
of foam [5]. the FARNNEL flat plate specimens and also literature
test results [6]. These data is then converted into true
stress and strain values as required by material model
4. Finite element modeling and simulation M-24, linear piecewise plasticity model of LS-DYNA3D
materials library. The selection of this model is recom-
The finite element simulation aspects for helmet im- mended by the LS-DYNA3D user manual for crashwor-
pact are presented here. At the beginning the simulation thiness applications and also used by other researchers
tools are described, then the helmet components model- for the same ABS material [20]. The results are shown
ing are presented and discussed. Afterwards, the com- in Table 1.
plete helmet model is verified against the helmet
standard and simulations are extended for the response 4.2. Finite element modeling of EPP foam liner
surface DOE requirements.
Simulation tools: Finite element simulation of struc- In order to model and validate this type of foam the
tural problems involves three basic steps. The first is experimental impact test results of Chou et al. [5] and
the pre-processing step; then the analysis step; and the the stress strain input data of Eq. (3) are used in this
post-processing step. In this study, the Hypermesh com- work. Chou et al. experimental test results involves the
puter code has been used for the first step. Hypermesh is impact of a hemispherical aluminum headform of
part of the hyperworks finite element package produced 4.5 kg weight on a cubical 63.3 kg/m3 EPP foam sample
by Altair Engineering. This program is a high perfor- with the dimension of 254 mm · 254 mm · 76 mm. The
mance finite element pre and post processor that enables impact speed is 6.7 m/s for the experimental test results,
the user to generate finite element models for engineer- which is very close to the 6 m/s of the Malaysian stan-
ing simulation and analysis. dards speed requirement. The small difference between
The analysis step involves the non-linear transient the two speeds is assumed to be negligible.
prediction for dynamic behavior. The LS-DYNA3D Foam Model Selection: The low-density foam model
has been used herein. Finally, the Finite Element Model M-57 of LS-DYNA3D material library is selected for
Builder (FEMB), GRAPH, and Hypermesh are used for modeling the EPP foam behavior. This selection is made
post processing the results. according to LS-DYNA3D user manual and also used
LS-DYNA3D is a large-scale non-linear finite ele- for polypropylene foam modeling by other researchers
ment package [18]. This code performs a non-linear [4,5]. Furthermore, this model is found to be the most
transient dynamic analysis of three-dimensional struc- suitable among the other LS-DYNA3D foam models
tures. Originally developed by DYNA3D-family of in terms of representing physical EPP foam behavior un-
computer codes at the Livermore National Laboratory,
LS-DYNA represents the commercial version of these
codes and is available through the Livermore Software Table 1
Technology Corporation [21]. LS-DYNA has a wide True stress and true strain of data for tensile test of ABS material used
in shell modeling
variety of analysis capabilities including a large number
of material models, a variety of contact modeling op- True strain 0 0.0174 0.0371 0.095 0.182 0.26
True stress 0 35.6 32.17 34.6 37.8 44.2
tions, a large library of beam, plate, shell, and solid ele-
ARTICLE IN PRESS
der compression and simplicity of input data. This mod- The good correlation can be clearly seen between the
el requires uniaxial stress–strain data and other param- experimental and the simulation results.
eters, which can obtained by standard laboratory tests,
unlike other foam models, which requires the dynamic 4.3. The helmet finite element modeling
pressure against volumetric strain experimental data
and other parameters that is currently not available The task here is to model the complete helmet–head-
and/or difficult to obtain. Furthermore, helmet impact form system and impact anvil. This model will be used
condition involves mainly a uniaxial compression load- for impact simulation of the helmet design verification
ing. This is also confirmed by Gilchrist, and Mills who and also fore response surface design analysis and opti-
had large contribution in helmet and foam studies [16]. mization. Some of the geometrical details, which are
Foam Modeling and Validation: Upon simulation tri- generally meant for other than mechanical performance
als of EPP foam using this model, it is found that the im- requirements such as aerodynamics and styling, are
pact stress–strain curve needs to be calibrated in order omitted in this study. This is also necessary because re-
to fit the experimental results. The most sensitive param- sponse surface analysis and optimization requires the
eter found to be the collapse stress, which is in some ref- generation of many complete helmet models with differ-
erences termed the yield stress. The slight tuning of this ent helmet parameters. Therefore building excessively
parameter has a little change on the overall shape of the detailed model for each simulation run will make this
experimental foam stress–strain curve, but it consider- job completely tedious and not practical to perform.
ably affects the simulation peak acceleration results. This computational helmet–headform model consists
The EPP foam is modeled using solid hexagon element of a hemispherical aluminum headform, the EPP foam
type with full integration points to avoid hourglassing liner and the ABS shell. The headform used has a hemi-
effects as recommended by the user manual for compo- spherical shape. Several materials can be used for head-
nents that undergoes large deformations [18]. The gen- form such as magnesium, aluminum, or hardwood (MS1
eral automatic contact type is used to model all 1996, and BS 6658:1985). As experimental test results
contact surfaces in this validation model. The contact for aluminum headform impact is currently available
stiffness has to be scaled up to avoid significant penetra- which is used for EPP foam model validation explained
tion of the element edges between the contacted parts. previously, it is used in the helmet–headform simulation
The element type selected for the headform is also solid model. The headform, and the foam liner are modeled
hexagonal type with single integration points. The mate- using hexagon solid element type of the LS-DYNA3D
rial of the headform is aluminum, which is modeled element library, while the shell is modeled using the
using the elastic material model M-1 of the LS-DY- thick shell element type. Table 2 shows modeling details
NA3D material library. The stiffness of the model is in- of the complete helmet system under consideration. The
creased to avoid headform elastic deformation, which helmet models are shown schematically in Fig. 4(a) and
may interfere with the foam energy absorption results. (b). Two types of anvils are used in helmet design verifi-
A rigid material can be selected, but the contact with cation. The first is circular flat with diameter of 125 mm,
elastic materials is more stable. Fig. 3(a) and (b) shows while the second is hemispherical shape with 50 mm ra-
the model used and the validation results, respectively. dius. These types of anvils are typical for almost all mo-
Table 2 the shape of the flat anvil impact curve for uniform shell
Finite element model details and liner deformation.
Component Material Element type No. of elements As shown, the model predicts lower peak acceleration
Headform Aluminum Solid 330 value for the hemispherical anvil impacts compared to
Foam EPP Solid 564 the flat anvil. This might be attributed to the ABS shell
Shell ABS Thick shell 564 local buckling, which led to more shell and liner energy
Flat Anvil Steel Solid 252
absorption contributions. This is a positive point in EPP
Total 1710 lined helmet design, as the helmet design for flat anvil
which is the most frequent impact shape [14] will be
more conservative for hemispherical and other similar
objects. However, this finding cannot be generalized
for all shapes as very sharp object will result in different
mode of helmet failure and the acceleration limit criteria
will not be applicable. This suggests that if a precise ef-
fect of the impactor shape is needed a further study on
this particular item might be performed.
In summary, the previous results proved that the use
of the EPP foam as a helmet liner is found to satisfy the
helmet standards in terms of the peak acceleration. The
general acceleration trend is similar to the EPS lined hel-
met. Therefore these results encouraged the authors to
extend the simulation work to cover the helmet design
analysis and optimization using the response surface
method. This method is explained in the next section.
Fig. 4. The finite element helmet models used in the simulation.
Fig. 5. The results of the helmet impact simulation using the EPP(65 kg/m3) foam. (a) Headform CG acceleration for helmet impact on flat anvil.
(b) Headform CG acceleration for Helmet impact on hemispherical anvil.
The function f(x) is normally chosen to be a low or- Once the model to characterize the response is se-
der polynomial, typically linear or quadratic. A qua- lected, then the coefficients b0. b1, bn will have to be esti-
dratic model may be represented mathematically as mated. Typically, a least squares method that minimizes
follows: the error residual is utilized for estimating the coeffi-
X X cients of the model equation [27].
y ¼ b0 þ bi xi þ bij xi xj ; ð5Þ
i i<j
Response surface DOEs: RSM has methods for select-
ing data points at which the response function should be
where bÕs are the coefficients that describe the response evaluated. The procedure of choosing a small set of opti-
function. The estimated coefficients of the fitted model mal points in the design space to fit the approximation is
will be denoted by b, for n variables, the quadratic mod- termed ‘‘design of experiments’’. The design points are
el equation (5) has a total of p = (n + 1)(k + 2)/2 chosen to maximize the predictive capabilities of the re-
coefficients. sponse function. In cases where there are significant
Response surface construction: If f(x) accurately de- noise (variance) errors, the points are chosen such that
scribes the process being modeled, e may be considered they minimize the variance error in the fitted coefficients.
as a random error, often assumed to have a normal dis- For problems with small numbers of design variables in
tribution with zero mean. In this case f(x) is an unbiased regularly shaped domains, selection schemes, called
estimate of y and e accounts for sources of variation not standard designs, are available.
accounted for by f(x), but which are still inherent to the In the present research the standard DOE scheme is
process such as noise and measurement errors. adopted. The former is the central composite experi-
ARTICLE IN PRESS
ment, and the later is the Box–Behnken designs. The fea- Table 3
tures of each schemes is highlighted in the forthcoming The response surface design levels
paragraphs and a selection of the most suitable scheme Level Foam Foam Shell
for this research is made. density (kg/m3) thickness (mm) thickness (mm)
Central composite designs: Central composite designs (1) Low 45 15 3
are often recommended when the design plan calls for (0) Medium 55 20 4
(+1) High 65 25 5
sequential experimentation because these designs can
incorporate information from a properly planned facto-
rial experiment. The factorial or ‘‘cube’’ portion and sign factors considered in this helmet design research
center points may serve as a preliminary stage where a are, the foam density, the foam thickness, and the shell
first-order (linear) model can be fitted, but still provide thickness. Their levels are estimated based on several
evidence regarding the importance of a second-order factors. Among those are the current helmet designs
contribution or curvature. Central composite designs with EPS foam liner, and some simulation trials with
usually have axial points outside the ‘‘cube’’ (unless EPP foam as a liner. The factors and their design levels
parameter is specified that is less than or equal to are shown in Table 3.
one). These points may not be in the region of interest, Response surface creation and analysis: After the de-
or may be impossible to run because they are beyond sign factors and their levels are set, the DOE matrix
safe operating limits. Central composite designs also al- based on the Box–Behnken method is created. The finite
low for efficient estimation of the quadratic terms in the element helmet model parameters are then varied
second-order model, and it is easy to obtain the desir- according to the DOE matrix settings. Crash simulation
able design properties of orthogonal blocking and rotat- is carried out for each design point (helmet model) and
ability. Orthogonally blocked designs allow for model the peak linear acceleration (response) is recorded and
terms and block effects to be estimated independently keyed into the design matrix as appropriate. When all
and minimize the variation in the regression coefficients. data are completed the response surface is analyzed.
Rotatable designs provide the desirable property of con- The full quadratic model is selected to analysis the re-
stant prediction variance at all points that are equidis- sponse surface. This model is selected because few vari-
tant from the design center, thus improving the quality ables are involved in this study and the selected model
of the prediction. will allow for the detection of even small curvature in
Box–Behnken designs: Box–Behnken designs are gen- the response surfaces. This is beneficial for more detailed
erally recommended when performing non-sequential investigation of the parameters effects. The output re-
experiments. That is, when planning to perform the sults are generated in the form of three dimensional re-
experiment once. These designs allow for efficient esti- sponse surface plots and contour plots. The discussion
mation of the first and second order coefficients. Because of these results is given in the next section.
Box–Behnken designs have fewer design points, they are
less expensive to run than central composite designs with 5.3. Response surface results and discussion
the same number of factors.
Also, Box–Behnken designs can prove useful if the The response surface design matrix with the simula-
safe operating zone for the design under consideration tion results keyed in is shown in Table 4 and the re-
is known. Box–Behnken designs do not have axial sponse surface graphs and contour graphs are shown
points, thus, all design points will fall within the safe in Figs. 6–11.
operating zone and also Box–Behnken designs ensure Each graph have the peak linear acceleration ‘‘ACC’’
that all factors are never set at their high levels as a response parameter plotted against two of the three
simultaneously. helmet design factors, which are the foam density
‘‘ROW’’, the foam thickness ‘‘T-foam’’ and the shell
5.2. The response surface solution thickness ‘‘t-shell’’.
Fig. 6 shows the peak linear acceleration in gÕs plotted
According to the above reasons, and due to fact that against the foam thickness and the foam density. As can
simulation results are constant response results and be seen, the curve has a smooth slope upward in the
either randomization or replication will not alter the re- foam density face, while it show a steeper downward
sponse. Therefore, the most economical design scheme curve for the foam thickness effect face. Investigating
which is the Box–Behnken design is adopted in this re- each parameter at a time, so considering the foam den-
search program. In the next paragraphs the procedure sity effect, the acceleration has a value of about 275g at
followed in the response surface solution is described. the lower foam density level of 45 kg/m3. Then, the
The response surface design factors and their levels: acceleration starts to smoothly increases as the foam
For the response surface to be created the design factors density increase until reaching to about 290g at the
and their levels need to be specified first. The three de- 65 kg/m3 end. This might give impression that if the
ARTICLE IN PRESS
T-foam
5 + 0 20
6 0 0 0
19
7 0 +
8 + 0 18
9 0 17
10 0 + 16
11 0
15
12 0 + +
13 0 45 55 65
14 0 0 0 Row
15 + 0 +
Fig. 7. The contour plot of the acceleration against the foam thickness
and the foam density.
255
245 65 kg/m3. If the acceleration needs to be kept below
235 230g, the foam thickness needs to be equal to or more
225 than 19.5 mm for foam density below 60 kg/m3. This
215 25
205
type of plot can be very useful in design as once one
20 T-foam parameter is specified the other parameter can be deter-
45
55 15 mined based on a given acceleration limit.
65
Row Therefore both graphs had explored a significant part
Fig. 6. The response surface of the acceleration against the foam of helmet design. It is now required to have the same
thickness, and the foam density, ROW: Foam density (kg/m3), T- investigation of each of the previous factors with the
foam: Foam thickness (mm), t-shell: Shell thickness (mm), Acc: Peak shell thickness parameter.
Headform CG acceleration (gÕs). Fig. 8 shows the peak acceleration plotted against the
foam density and the shell thickness. The significant ef-
fect of the foam density is evident as compared to the
peak acceleration needs too be kept below 300g it will be shell thickness. The shell thickness seams to have very
necessary to keep the foam density below the 65 f kg/m3. little effect on the peak acceleration. This is clear from
This could be further investigated on the other graphs. the shallow curvature of the response surface toward
Considering the foam thickness effects, as can be seen, increasing the shell thickness. However, both the foam
the acceleration decreases as the foam thickness in- density and the shell thickness design values did not
crease. The amount of decrease is quite considerable show any risk of approaching the helmet standard injury
as compared to the foam density, which shows that limit of 300g. This can be further investigated from the
the foam thickness effect is more important for lowering contour plots, which is presented hereafter.
the acceleration level. However, for the selected thick- Fig. 9 shows the contour plot of the peak linear accel-
ness range, the observation is that there is no curve eration against the foam density and the shell thickness.
inversion (rise portion) as the foam thickness increase. As shown, from this curve less information can be ob-
This is expected as theoretically, by increasing the foam tained about the foam density effects. However, the shell
thickness, the acceleration will keep in decreasing, and thickness effects are more noticeable. This can be seen
no any reason will return the acceleration to the increas- from the 232g and the 252g constant acceleration lines.
ing stage again. For instance, if the 4 mm shell thickness with the corre-
Furthermore, to complete the analysis of these sponding foam density of 52.5 kg/m3, which is the
parameters effect on the acceleration, it is worth to in- curve inversion point, is considered as a reference point
ARTICLE IN PRESS
Surface Plot of Acc Fig. 10 shows the response surface graph of the peak
linear acceleration against the foam thickness and the
shell thickness. As can be seen, the acceleration in-
creased from about 220g to about 280g by decreasing
the foam thickness from 25 to 15 mm. However, the
shell thickness effect is of minor significance. This is evi-
250
dent from the small increase of the peak acceleration by
increasing the shell thickness from 3 to 5 mm. These ef-
240 fects can also be observed from Fig. 11, which shows the
Acc
260
4
250
Acc
240
230
220 5
210 4
3 15
3 t-shell
20
45 55 65 25
T-foam
Row
Fig. 10. The response surface of the acceleration against the foam
Fig. 9. The contour plot of the acceleration response against the shell
thickness, and the shell thickness.
thickness and the foam density.
in the 232g line. Then, by decreasing the shell thickness Contour Plot of Acc
to 3 mm, the density needs to be increased to 55 kg/m3 5
230
to lie on the 232g acceleration line. On the other hand, 240
increasing the shell thickness to 5 mm needs that the 250
foam density be increased to 55 kg/m3 to achieve the 260
270
same result. From the previous it cab be concluded that
280
the 4 mm and the 52.5 kg/m3 are the optimum setting
t-shell
4
for the 232g line. However, the situation is different
for the 252g acceleration line, which is almost straight
line. In this later line, changing the shell thickness dose
not seem to have any effect on the foam density. So by
comparing the two curves, we can conclude that the
3
shell thickness effect is more evident for the lower den-
sity designs, than the higher ones. In general, this curve 15 16 17 18 19 20 21 22 23 24 25
needs to be investigated in conjunction with other re- T-foam
sponse and contour graphs to have a more comprehen- Fig. 11. The contour plot of the acceleration against the foam
sive visualization. thickness and the shell thickness.
ARTICLE IN PRESS
Table 5
The D-optimal design matrix (Box–Behnken design)
Order Run order Row T-foam t-shell Peak acceleration (g)
14 1 55 20 4 235
4 2 65 25 4 240
7 3 45 20 5 223
3 4 45 25 4 204
2 5 65 15 4 296
15 6 55 20 4 235
10 7 55 25 3 221
6 8 65 20 3 245
9 9 55 15 3 282
11 10 55 15 5 266
5 11 45 20 3 220
12 12 55 25 5 224
1 13 45 15 4 280
13 14 55 20 4 235
8 15 65 20 5 254
Table 6
The selected D-optimal design points
Foam Foam Shell Peak
density (kg/m3) thickness (mm) thickness (mm) acceleration (g)
55 15 5 266
55 20 4 235
Fig. A.2. The dynamic stress strain curves for PP and rigid PU foam
65 20 5 254
of various densities [24].
ARTICLE IN PRESS
[24] Mills NJ. Impact response. In: Halyard NC, Cunningham A, [28] Montogmery DC, Runger GC, Hubel NF. Engineering statistics;
editors. Low density cellular plastics. Physical Basis of Behavior; 2001.
1994. [29] SolidGold Helmet Sdn.Bhd., Balakong, Selangor Darul Ehsan,
[25] Mills NJ. Accident investigation of motorcycle helmets. J Inst Malaysia.
Traffic Accident Investigations, Aut. 1996;5:41–56. [30] Vetter L, Vanderby R. Influence of materials and structure on
[26] Mills NJ, Gilchrist G. Shear and compressive impact of PP bead performance of a football helmet. Polym Eng Sci 1987;27(15):1113–20.
foam. Cell Polym 1999;18:157–74. [31] Yetham LA, Godfreg PM, Chinn PB. Materials for motorcycle
[27] Montgomery DG. Design and analysis of experiment. New crash helmets: a finite element parametric study. Plastics, Rubber
York: John Wiley and Sons Ltd.; 1991. Compos Process Appl 1994(22):215–21.