A Fracture Mechanics Approach To Durability Calculations For Adhesive Joints

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 11

INTRODUCTION

Adhesive bonding is increasingly used in the development


of lightweight vehicle bodies. In addition to improving
structural rigidity, it provides a means of joining high
strength steels, aluminium alloys and dissimilar materials,
with good durability against fatigue compared to more
traditional methods (spot-welding and seam-welding). As
with any manufacturing technology, if full advantage is to be
taken of its benefits, analytical methods are required in order
to be able to evaluate and optimize the durability of the
resulting structure. Without these analytical methods, and in
an environment where development times are very short,
release of product with unnecessary over-design and/or
unquantified safety factors is inevitable.
Ideally, adhesive joints should be designed so that they
are loaded primarily in shear (rather than peel or cleavage
modes), and in addition to having a uniform thin layer of
adhesive, fillets are highly desirable in order to minimize
stress concentrations at the periphery of the joint. In contrast,
adhesive joints in mass-produced automotive vehicle bodies
will in general experience complex loadings including
elements of shear and peel, and may be underfilled (primarily
for cosmetic reasons). Under sufficient static or cyclic
loading, such joints are likely to fail by crack propagation
through the joint. Cracks may be interfacial or cohesive, or a
2012-01-0731
Published 04/16/2012
Copyright 2012 SAE International
doi:10.4271/2012-01-0731
saematman.saejournals.org
A Fracture Mechanics Approach to Durability Calculations for
Adhesive Joints
Peter Heyes
HBM UK, Ltd.
Gunnar Bjrkman
Volvo Technology
Andrew Blows and Tim Mumford
Jaguar/Land Rover Cars
Paul Briskham
Coventry University
ABSTRACT
Effective use of adhesive bonding in automotive vehicle bodies requires analytical methods for durability, so that
potential fatigue problems and unnecessary overdesign may be eliminated before the physical prototype stage and release
of product with unquantified safety factors avoided.
This paper describes a fracture mechanics-based method for predicting the durability of adhesive joints, based on work
previously carried out at Volvo [1]. The method requires relatively modest modifications to a typical vehicle body FE
mesh. Adhesive bonds are represented by bar elements around the periphery of each bond. Grid point forces from shell
elements adjacent to the adhesive bond are recovered and used to determine line forces and moments at the edge of the
glued flange. These forces and moments are then transferred to an analytical sandwich model of the joint. This enables
approximate calculations of the strain energy release rate (or rather the equivalent J-integral) to be made for assumed small
cracks at the edge of the adhesive. Analytical results are compared with calculations based on detailed FE meshes for
typical test specimen geometries, and with physical test data. J-integral values calculated for different geometries and
loadings are shown to be able to correlate the durability of joints, particularly at longer lives, enabling useful estimations of
joint durability to be made.
CITATION: Heyes, P., Bjrkman, G., Blows, A., Mumford, T. et al., "A Fracture Mechanics Approach to Durability
Calculations for Adhesive Joints," SAE Int. J. Mater. Manf. 5(1):2012, doi:10.4271/2012-01-0731.
____________________________________
215
Downloaded from SAE International by University of Michigan, Monday, September 15, 2014
mixture of these failure modes may occur, depending on
factors such as the choice of adhesive, surface pretreatment
and surface cleanliness prior to adhesive application. See
Figure 1.
Figure 1. Adhesive joints: Figure 1(b) is more typical of
an automotive vehicle body.
This paper describes a method for evaluating the
durability of such adhesive joints in vehicle bodies, based on
work previously carried out at Volvo by Bjrkman [1]. It is
based on the following premises:
Because of the nature of the joints and their manufacturing
process, their durability is best described using a fracture
mechanics approach.
Stress intensity factors are not particularly useful for cracks
in inhomogeneous bodies such as an adhesive joint, or for
interfacial cracks. The loading on a crack is therefore
evaluated using the strain energy release rate, or rather since
we are considering only limited plasticity around the crack
tip, we can use the equivalent J-integral.
Any calculations must be capable of being carried out on a
typical body durability model, with minor modifications only.
In the body CAE process, detailed FE modelling of the type
normally used to carry out J-integral calculations is out of the
question, so an approximate method is required.
Several factors conspire against making accurate life
estimations - the imprecise nature of the manufacturing
process, rather flat SN curves (steep crack growth curves) for
the adhesive, and the fact that the J-integral is a scalar
quantity. For this reason, in the interests of making some
useful calculations, the method described here is limited to
the determination of maximum J-integral values and
comparison with a threshold value. Finite life predictions are
not attempted.
The calculation process is illustrated in Figure 2.
Figure 2. Summary of calculation process
The basic steps of the method are as follows:
1. A global finite element analysis of the structure is
carried out. A simple, relatively coarse representation of the
joints is used in which beam elements represent the adhesive
joining the flanges.
2. Line forces and moments along the edges of the
flanges are derived from the FE results, combined with
applied loading histories.
3. The line forces and moments are applied to an
analytical sandwich model of the bonded flange, and used
to calculate the strain energy release rate G (actually the J-
integral) based on an initial small crack in the adhesive.
4. At each point along the edge of the flange, the
calculated maximum J-integral during the loading history is
compared to the threshold required for crack growth G
th
to
determine whether or not the joint is likely to fail.
The method does not calculate fatigue life, but in addition
to the maximum value of the J-integral for a given flaw size,
an estimated reserve factor may be calculated.
THEORETICAL BACKGROUND
MODELLING GUIDELINES AND FE
ANALYSIS
FE models for adhesive joint durability calculations
require only minor changes from typical automotive body
models. The modelling guidelines may be summarized as
follows:
Models should be created using linear shell elements
positioned at the mid-planes of the sheet metal parts.
Heyes et al / SAE Int. J. Mater. Manf. / Volume 5, Issue 1(April 2012) 216
Downloaded from SAE International by University of Michigan, Monday, September 15, 2014
Each adhesive bond should be represented by beam
elements around the periphery of the bond, connecting the
shell meshes of the joined parts. Typically beams will have a
5mm circular cross-section, with modulus 3000 MPa and
Poisson ratio 0.4.
Bonded flanges require more or less congruent and parallel
meshes so that the beams are normal to the flanges.
Beam elements should be as close as possible to the
position of the edge of the adhesive in the bond, which means
that ideally the bend radius adjacent to the joint is represented
by at least one row of shell elements as in the preferred
method illustrated in Figure 3.
Figure 3. Mesh cross section - preferred modelling
practice.
Any offset between the position of the beam elements and
the actual edge of the adhesive (e.g. if the acceptable
method in Figure 3 is used) will result in errors in the forces
and particularly the bending moments at the edge of the joint.
The method compensates for this when it estimates the
bending moments at the crack tip, but any offset should be
minimized for best accuracy.
The purpose of the FE analysis is to provide information
that can be used to derive time-histories of the J-integral for a
small crack at various points along the edge of the adhesive
bond. An intermediate step is the derivation of time-histories
of force and moment per unit length (line forces and
moments) applied to the sandwich (flanges plus adhesive)
by the adjacent elements. These time histories can be
generated by linear superposition (e.g. by combining static
unit loadcase results and corresponding load-time histories),
direct from FE time-steps, or from a transient analysis,
whether a direct transient or using a modal superposition
approach. In any case, the data required from the FE analysis
are the nodal forces and moments from the shell elements
adjacent to the flanges.
At each calculation point, the nodal forces and moments
in the shells adjacent to the bonded joint are used to
determine line forces and moments on the upper and lower
elements of the joint, in a local co-ordinate system.
Figure 4. Calculation points and local co-ordinate
system
The method for determination of the line forces and
moments is based on that originally proposed by Fermr et al
[2] for determination of structural stresses at the toe of a
seamweld, and refined by Heyes in nCode DesignLife [3].
J-INTEGRAL CALCULATION
The concept of using a sandwich model to calculate J-
integrals and predict fracture has been described by Fernlund
et al [4]. This idea was developed further by Bjrkman [1],
working at Volvo, who extended its application by
transferring loads from a shell model, and critically,
improved the accuracy of the calculation through the
introduction of a distortion term, as outlined below.
The basic premise of the method is that the strain energy
release rate G associated with a small crack in the periphery
of an adhesive bonded joint in a vehicle body can be
determined by transferring the line forces and moments along
the edge of the representation of the bond in a shell FE model
of the body to an anlytical sandwich model of the joint,
which is used to calculate G. In practice, what is actually
calculated is the equivalent path-independent J-integral as
described by Rice [5]. There are two assumptions implicit
here:
1. In order for the strain energy release rate G to be
equivalent to the J-integral, any yielding must be limited to a
small region around the crack tip. This is reasonable when
Heyes et al / SAE Int. J. Mater. Manf. / Volume 5, Issue 1(April 2012) 217
Downloaded from SAE International by University of Michigan, Monday, September 15, 2014
considering the endurance domain; widespread plasticity
would give rise to low-cycle fatigue failure.
2. Cracks are not large enough to cause significant re-
distribution of load around the adhesive joints.
For convenience, the J-integral is evaluated along a path
just before and just after the crack front, as illustrated in
Figure 5.
Figure 5. Transfer of line forces and moments to
sandwich model and J-integral evaluation path
The basic inputs to the analysis are:
: the line forces and moments on the
upper shell
: the line forces and moments on the
lower shell
These are used to estimate the cross sectional line forces
and moments (N,M,T) in the upper and lower parts of the
sandwich just before the position of the crack front, and in the
sandwich just beyond the crack front, as illustrated in Figure
6. The effective total crack depth a is the sum of the distance
(offset) from the position of the beam elements in the shell
model to the edge of the adhesive plus the actual flaw depth.
Figure 6. Sandwich model schematic



The total strain energy release rate is then computed from
the J-integral. The J-integral is evaluated along a path which
has non-zero contributions from three sections, corresponding
to the top sheet, and bottom sheet just behind the crack front,
and the sandwich just ahead of the crack front. These are
denoted by subscripts 1,2 and 0 respectively in the following
equations.
First the modulus and effective thickness of the sandwich
are defined:
(1)
(2)
Then additional moduli, for i = 1,2:
(3)
(4)
and for i = 0:
(5)
(6)
Then the distance from the bottom to the centroid of the
sandwich
(7)
Second moment of area for i = 1,2
(8)
and for i = 0:
Heyes et al / SAE Int. J. Mater. Manf. / Volume 5, Issue 1(April 2012) 218
Downloaded from SAE International by University of Michigan, Monday, September 15, 2014
(9)
The distortion (slope) of the beam elements makes a
significant contribution to the strain energy release rate, and
hence the J-integral. The rotation of the beam cross sections
is estimated from:
(10)
for i=1,2, and
(11)
The slope constants K are empirical functions of the sheet
and adhesive thicknesses and moduli. For the current work,
these functions have been determined based on analytical
experiments with sheet thicknesses between 1 and 3 mm for
steel and aluminium sheets.
Now we can calculate the J-integral
(12)
and finally
(13)
SOFTWARE IMPLEMENTATION
The method described in the previous section has been
implemented in the fatigue analysis software package nCode
DesignLife. This allows the user to create an analysis
process, within a graphical user interface, that can estimate
the J-integral as a function of time for small cracks at points
all along the edge of adhesive bonds defined in a finite
element model. Loadings can be defined by linear
superposition of static loading cases, by direct or modal
transient calculations, or as a duty cycle comprising any
combination of these loading types. The output from the
analysis is the maximum value of the J-integral at each
calculation point over the period of loading. The result is
assigned to adjacent elements for ease of post-processing.
The full time-history of the J-integral may also be recovered
if required.
In addition, an estimated reserve factor, based on the ratio
of J
max
and the threshold strain energy release rate G
th
may
also be calculated. Since J and G scale as the square of the
applied load, the reserve factor has been defined as follows:
(14)
A simple analysis process is illustrated in Figure 7 below:
Figure 7. Simple analysis process in nCode
DesignLife, based on coach-peel test specimen
VALIDATION
The validation of the method was carried out in two
stages:
1. The ability of the method to calculate J-integrals on the
basis of grid point forces recovered from a coarse FE model
was tested against conventional fracture mechanics
calculations using a fine FE mesh.
2. The usefulness of the J-integral as a means of
correlating the durability of various adhesive joints was
evaluated using physical test data.
J-INTEGRAL CALCULATION
The J-integral calculation was validated by comparing
values calculated using the approximate method of Bjrkman,
based on simple structural models, with results computed
using ABAQUS models with detailed FE meshes. The
geometries chosen are those of the simple lap-shear and
coach-peel specimens used in the fatigue test program. The
Heyes et al / SAE Int. J. Mater. Manf. / Volume 5, Issue 1(April 2012) 219
Downloaded from SAE International by University of Michigan, Monday, September 15, 2014
specimens are made from 2 and 3 mm aluminium sheet, in all
possible thickness combinations, bonded with an epoxy
adhesive. In order to ensure a consistent bond width and
thickness, and to prevent the formation of any kind of fillet,
the bonded area in all specimens was delineated using PTFE
tape. The result in all cases is a bonded area 20 mm wide in a
nominal flange width/overlap of 23 mm, and bond thickness
of 0.25 mm.
The specimen geometries are illustrated in Figure 8.
Figure 8. Coach peel and lap-shear specimen geometries
The simple structural models of the specimens are based
on the preferred modelling guidelines described earlier. A
couple of examples are illustrated in Figure 9.
Figure 9. FE models of lap-shear and coach peel
specimens

The contour integral (J-integral) calculations in ABAQUS
were carried out using detailed models constructed from 2D
plane strain elements, representing a section through the
symmetry plane of each specimen. Crack depths of 0.5 - 3.5
mm were considered. An example of the sort of mesh used is
illustrated in Figures 10 and 11. Calculations were made for
cracks through the middle of the adhesive layer (cohesive
failure) and at the interface between the aluminium and the
adhesive (interfacial failure). For asymmetric specimens, both
interfaces were considered.
Calculated J-integrals from the detailed models for the
defined crack depths are compared with results calculated
using the approximate method implemented in DesignLife,
based on NASTRAN models. To avoid end effects, the
approximate J calculations were taken from the middle of the
specimen. The results are shown in Figures 12,13,14,15,16,17
for all the different test configurations.
Figure 10. Detailed 2D model of coach peel specimen
with 0.5 mm crack
Figure 11. 2D model - close up of crack tip area
(deformed plot) - cohesive failure
Heyes et al / SAE Int. J. Mater. Manf. / Volume 5, Issue 1(April 2012) 220
Downloaded from SAE International by University of Michigan, Monday, September 15, 2014
Figure 12. Approximate vs Detailed J-integral
calculations. Lap-shear 2-2 mm specimen.
Figure 13. Approximate vs Detailed J-integral
calculations. Lap-shear 3-3 mm specimen.
Figure 14. Approximate vs Detailed J-integral
calculations. Lap-shear 2-3 mm specimen.
Figure 15. Approximate vs Detailed J-integral
calculations. Coach peel 2-2 mm specimen.
Figure 16. Approximate vs Detailed J-integral
calculations. Coach peel 3-3 mm specimen.
Figure 17. Approximate vs Detailed J-integral
calculations. Coach peel 2-3 mm specimen.

Heyes et al / SAE Int. J. Mater. Manf. / Volume 5, Issue 1(April 2012) 221
Downloaded from SAE International by University of Michigan, Monday, September 15, 2014
CORRELATION OF FATIGUE TEST
DATA
The specimens described above were fatigue tested in
tension-tension at constant amplitude, with load ratio R = 0.1
and in load control. Displacement-time histories were
recorded for all specimens.
Additional hybrid riv-bonded and weld-bonded joint
specimens (combinations of self-piercing rivets - SPR - or
spotwelds with bonding) were also tested. The locations of
the rivets or spotwelds in these joints are illustrated in Figure
8. These hybrid configurations are not the main subject of
this paper and while they are mentioned here, the findings
from tests on these specimens will be described in more detail
in further publications.
Load-life curves for all 6 bonded-only specimen
geometries tested (lap-shear and coach-peel specimens in all
combinations of 2 and 3 mm sheet) are summarized in Figure
18.
Figure 18. Load-life curves for bonded specimens
The next step was to evaluate the J-integral as a means of
correlating the data from different test specimen geometries.
For each specimen geometry and load level, the J-integral at
the mid-point of the bond was calculated using the
approximate method, based on simple NASTRAN models,
assuming an initial crack size of 0.4 mm. The choice of 0.4
mm for the initial defect size was suggested by Bjrkman [1]
based on inspection of test specimens, and correlation of test
data. Ideally the choice of initial defect size should be based
on measurements from production parts, but in practice the
value of the calculated J-integral is not very sensitive to small
variations. For example, for a coach peel specimen, initial
crack sizes of 0.2 mm and 0.8 mm result in a reduction of 5%
and an increase of 11% in the calculated J-integral
respectively, compared to a 0.4 mm crack. Lap-shear
specimens are less sensitive.


In Figure 19, data from the same tests as those in Figure
18 are plotted with the calculated J-integral as the vertical
axis. Note that the J-N curves converge in single scatter band,
particularly in the endurance domain (10
6
10
7
cycles).
Figure 19. J-life curves for bonded specimens
The good correlation of test data with J-integrals based on
a small assumed initial crack size, particularly in the
endurance (long-life) domain, encourages the idea that these
calculations might provide a useful basis for evaluating the
durability of real structural joints with a variety of sheet
thicknesses and loading types. The degree of scatter,
approximate nature of the J-calculations and relatively flat
curves indicate that we cannot expect to make realistic finite
life predictions, particularly for mass-produced parts.
However we should be able to predict whether or not the
loading on a joint is sufficient to exceed the crack growth
threshold for the adhesive.
One other matter should be touched upon here, and that is
the fact that many of the joints in an aluminium car body are
hybrid joints, for example combinations of adhesive and self-
piercing rivets. In such riv-bonded joints, the mechanical
fasteners are responsible for very little load transfer while the
adhesive is intact, and we would expect the presence of a
rivet to make very little difference to the J-integral at the edge
of the adhesive bond (although the riveting process will
inevitably affect the bonded area and uniformity of the bond
line thickness). This view is supported by Figure 20 in which
J-N data (solid circles) for riv-bonded coach peel (T-peel)
specimens made with 3 mm sheets are superimposed on the
bonded-only data from Figure 19.
Heyes et al / SAE Int. J. Mater. Manf. / Volume 5, Issue 1(April 2012) 222
Downloaded from SAE International by University of Michigan, Monday, September 15, 2014
Figure 20. Comparison of bonded and riv-bonded joint
durability, based on J-integral
At higher load levels, the adhesive cracks quite quickly,
but there is significant residual life in the joint due to the
resilience of the SPR, leading to a relatively steep slope
compared to the pure bonded joints. In the endurance domain,
the durability is dominated by the fatigue threshold of the
adhesive, and the data appears to converge with that of the
pure bonded joints.
APPLICATION
In order to test the usefulness of the method implemented
in the DesignLife software, it was necessary to apply it to a
realistic example, for which the Jaguar XJ body-in-white
(un-glazed) was selected. Considerable difficulty was
experienced in generating any fatigue failures in the structural
joints of the body during rig testing, and some
experimentation was required before a set of loads and
constraints could be identified capable of causing the joint
fatigue failures required to validate the method. The resulting
experimental test set-up is illustrated in Figure 21.
The body is constrained at the B-pillars and adjacent sills,
and loaded in torsion using 4 actuators acting at the shock
towers. The variable amplitude (VA) load signal for the
fatigue testing is derived from a representative surface of the
durability test circuit. The forces in the shock tower signals
were manipulated to give a pure torsional loading on the car
body. The signals for the actuators were scaled up to a high
level in order to produce fatigue failures within the required
test window. The VA signal was also modified to have a
constant frequency of 5Hz in order to avoid any significant
dynamic response in the structure.
Figure 21. Test rig with Jaguar XJ body-in-white
The fatigue test was simulated using a typical FE
durability model of the XJ body-in-white (BIW), lightly
modified in accordance with the modelling guidelines for the
approximate J-integral calculation method. Rivets were not
explicitly included in the FE model. The same VA amplitude
loading was applied to the model as used in the physical test,
and safety factors calculated based on a threshold strain
energy release rate (J-integral) of 100 J/m
2
and an initial
crack size of 0.4 mm. The analysis highlighted a joint in the
region of the rear seat back to C-pillar connection, on both
sides of the vehicle, as having a safety factor < 1.0.
Figure 22. Fatigue test simulation results and
corresponding cracking on test




Heyes et al / SAE Int. J. Mater. Manf. / Volume 5, Issue 1(April 2012) 223
Downloaded from SAE International by University of Michigan, Monday, September 15, 2014
On test, fatigue cracking was observed to occur in this
joint on both sides of the tested vehicle bodies. An example
from the left hand side of the vehicle is marked in Figure 22
(inset). Although the condition of the underlying adhesive in
this joint cannot be seen, it is surmised that the adhesive must
have failed in order for the SPR to experience the significant
fatigue loading required to generate this crack. Cracking of
the adhesive is more clearly visible in an example of the
corresponding joint on the right hand side of the body, as
illustrated in Figure 23.
Figure 23. Adhesive cracking in rear seat back to C-
pillar connection (top right of figure)
SUMMARY
This paper has described an approximate method for the
calculation of the J-integral associated with small cracks in
car body adhesive joints. The FE modelling guidelines
require only modest modifications to typical body durability
models. Investigations into the usefulness of this method
have had the following positive results.
1. The method has been shown to match J-integrals
calculated by explicit and detailed modelling of cracks to a
reasonable degree of accuracy (in the context of body
durability calculations).
2. The calculated J-integral for a small crack in the edge of
an adhesive joint has been shown to provide a useful means
of correlating the fatigue strength of different joints in shear
and peel loading in the endurance domain.
3. The method shows promise as a means of identifying
critically loaded adhesive and hybrid joints in vehicle bodies.
The method has a number of limitations.
1. Because the J-integral is always a positive scalar, it is not
suitable for cycle counting or making finite life predictions.
Having said that, the making of finite life predictions for the
type of joint covered by this paper may not be a realistic aim.
2. There is no consideration of the effect of load ratio.
3. Because the sandwich model is linear (no contact), the
calculated J-integrals do not distinguish, for example between
tensile and compressive loading on a peel joint.
In addition to exploring these issues, there is scope for
further work in a number of other areas.
1. Exploration of the effects of temperature, environment,
age and surface treatment on the durability of adhesive joints.
2. Improved modelling guidelines: the guidelines as
described lead to relatively high stiffness at the corners of a
bonded area, and the recommended beam element properties
are somewhat arbitrary.
3. Improved meshing tools. Experiments with a mesh
independent approach have so far had rather poor results.
4. Hybrid joint modelling
REFERENCES
1. Bjrkman, G., Global modeling and life prediction of adhesive bonded
structures, Volvo Engineering Report, ER-540039, 1999.
2. Fermr, M., Andrasson, M., and Frodin, B., Fatigue Life Prediction of
MAG-Welded Thin-Sheet Structures, SAE Technical Paper 982311,
1998, doi: 10.4271/982311.
3. Heyes, P., DesignLife theory guide, HBM UK Ltd, 2010 (v7.0).
4. Fernlund, G., Papini, M., McCammond, D. and Spelt, J. K., Fracture
load predictions for adhesive joints, Composites Science and
Technology, Vol. 51, 1994, pp. 587-600.
5. Rice, J., A Path Independent Integral and the Approximate Analysis of
Strain Concentration by Notches and Cracks, Journal of Applied
Mechanics, Vol. 35, 1968, pp. 379-386.
ACKNOWLEDGMENTS
The testing and software implementation work described
in this paper was carried out as part of the Bonded Car
collaborative research project, the partners being Jaguar Land
Rover, HBM UK Ltd (nCode), Coventry University,
Warwick University, Innoval Technology, Henrob and Stoke
Golding Applied Research. The project was co-funded by the
Technology Strategy Board's Collaborative Research and
Development programme, following an open competition.
The Technology Strategy Board is an executive body
established by the UK Government to drive innovation. It
promotes and invests in research, development and the
exploitation of science, technology and new ideas for the
benefit of business - increasing sustainable economic growth
in the UK and improving quality of life.
The theoretical basis of the work presented here was
developed as part of an earlier project conducted by the
Volvo Group, whose contribution is gratefully acknowledged.
DEFINITIONS/ABBREVIATIONS
h
1
the thickness of the top shell
h
2
the thickness of the bottom shell
h
a
the thickness of the adhesive layer
E
1
modulus of sheet 1 (upper sheet)
v
1
Poisson ratio of sheet 1
Heyes et al / SAE Int. J. Mater. Manf. / Volume 5, Issue 1(April 2012) 224
Downloaded from SAE International by University of Michigan, Monday, September 15, 2014
E
2
modulus of sheet 2 (lower sheet)
v
2
Poisson ratio of sheet 2
E
a
Modulus of adhesive
v
a
Poisson ratio of adhesive
Heyes et al / SAE Int. J. Mater. Manf. / Volume 5, Issue 1(April 2012) 225
Downloaded from SAE International by University of Michigan, Monday, September 15, 2014

You might also like