JOURNAL OF MATERIALS SCIENCE 18 ( 1 9 8 3 ) 1810-1826

A tensile testing technique for fibre-reinforced

composites at impact rates of strain
J. H A R D I N G , L. M . W E L S H
Department of Engineering Science, University of Oxford, Parks Road, Oxford, UK

A brief review is given of techniques which have been employed in attempts to determine
the mechanical properties of composite materials under tensile impact loading. The
difficulties encountered in the design of a satisfactory tensile impact testing machine for
composite materials are discussed and a new method, using a modified version of the
standard tensile split Hopkinson's pressure bar (SHPB), is described. Dynamic stress-
strain curves for unidirectionally-reinforced carbon/epoxy composite, in which failure
occurs in less than 30/~sec at a mean strain rate of about 400 sec -1, are presented and
their validity is established. An extension of the technique to allow the testing of woven-
roving reinforced glass/epoxy composites is described and dynamic stress-strain curves
obtained for which the times to failure approach 100/~sec and the average strain rate is
of the order of 1000 sec -1. Comparative stress-strain curves at low and intermediate
rates of strain are obtained and the effect of strain rate, over about 7 orders of magni-
tude, on the tensile modulus, and strength, fracture strain and energy absorbed in frac-
turing is determined. The limitations of the technique are discussed.

1. Introduction the split Hopkinson's pressure bar (SHPB) tech-

The need for a full characterization of the nique. In recent years this technique has been
behaviour of fibre-reinforced composites under adapted to the testing of composite materials in
dynamic loading conditions has prompted compression [8,9] and torsion [10] while a
numerous investigations in recent years [1-3]. high-speed punch version of the SHPB has been
Nevertheless, because of the experimental difficul- used [11] to determine the resistance to dynamic
ties involved few reliable data are available. Much perforation.
of the work that has been done has employed the The greatest difficulty has been experienced,
instrumented Charpy test in which a notched however, in obtaining reliable data for impact
beam specimen is subjected to impact bending tension. Tensile strain rates of about 10sec -1
[4, 5]. Although load-time records obtained from have been obtained in an apparatus using a gas-
the instrumented tup may be used to estimate the driven piston [12] while rates approaching 500
energy absorbed in the various stages of the fractur- sec -~ were achieved when an explosive charge was
ing process, stress wave reflections and the com- used to drive the impacting head [2]. In both
plex geometry of the specimen inhibit any funda- cases high-speed photographic techniques were
mental analysis of the material response and its required to monitor the strain and, in the latter
dependence on loading rate [6]. A need for the investigation, slipping at the clamps was a major
development of tests covering a wide range of problem. Drop-weight techniques have been used
loading rates, up to impact, in uniaxial tension by several investigators [ 13, 14], strain being deter-
and compression and in pure shear has been clear mined from strain gauges attached directly to
for some time [7] and has recently been restated the specimen. This method of strain measurement
[61. is very accurate up to the point at which the
For isotropic materials, tests at the highest gauges fail but is unsatisfactory for materials
rates of loading are frequently performed using which exhibit surface damage prior to failure. It
1810 0022-2461/83 $03.00 + .12 9 1983 Chapman and Hall Ltd.
is also clear that the drop-weight technique, as strain determined from strain gauges mounted on
developed in these investigations, is not free of the specimen and those determined from the
stress wave reflections in the load cell [3]. These Hopkinson-bar analysis.
are superimposed on the specimen stress-time The present paper describes the development
response. and validation of this technique and its application
Some of the difficulties have been overcome in to the tensile impact testing of uniaxially-reinforced
a technique proposed by Daniel et al. [15] where carbon fibre/epoxy composite. The behaviour
the specimen, in the form of a ring or shell, is observed is compared with that found in tests
loaded by means of an internal pressure pulse. In performed at lower rates. The technique is then
theory the specimen is stressed uniformly at all extended to allow the testing of woven-roving
times so stress wave reflections do not arise and reinforced glass fibre/epoxy composite where
problems associated with end effects and the fix- the strains to fracture are as much as a factor of
ing of the specimen to the loading bars are elimi- 10 greater than for the CFRP (carbon fibre rein-
nated. Nevertheless several major disadvantages forced plastic) material and results are presented
remain. The specimen experiences a continuously for tests on two orientations of specimen at rates
decreasing strain rate throughout the test. The from quasi-static to about 1000 sec -1 .
state of stress in the ring specimen is far from
simple including, as well as membrane stresses, a 2. Experimental details
pressure induced radial stress at the inside face In the original compression version of the Hopkin-
and shear stresses resulting from any variation in son-bar the specimen is sandwiched between two
the dynamic load along the axial direction. Also, elastic loading bars and the incident, reflected and
the practical difficulty of making the specimens transmitted stress waves are determined from
and actually performing the tests limits its value as strain gauges attached to these loading bars on
a technique for the rapid evaluation of a large either side of the specimen. In the standard tensile
number of materials. version of the apparatus [20] the input loading
The only previous application of the Hopkinson bar becomes the weighbar tube within which the
bar technique to impact tension testing of com- output, or inertia, bar slides freely, the specimen
posites is in the work of Kawata et al. [16]. In connecting the two at the yoke, see Fig. la. In a
their machine, which works on the bar-block specimen test strain gauges on the inertia bar
principle, a cylindrical tensile specimen with a monitor the transmitted stress wave. To determine
screw fixing is used. Accurately coaxial impact the input conditions, however, a separate "elastic"
is required on the impact block to minimize bend- test is required, performed under identical impact
ing stresses in the specimen. Three different conditions but with the specimen and the inertia
techniques are used to determine the velocity of bar replaced by an elastic bar. For tests on ductile
the block, which is assumed also to be that of the metal specimens this enables the dynamic plastic
impact end of the specimen gauge section. How- strain and strain rate to be determined with reason-
ever, in view of the low values of strain being able accuracy using specimens of a gauge length
measured and the lack of an independent check, short enough for the attainment of stress equili-
some doubt must remain regarding the accuracy brium across the specimen to be assumed from an
of the resulting strain determination. early stage in the test. This technique does not,
Very recently another tensile impact test [17] however, allow for an accurate determination of
using the Hopkinson bar technique has been strain during the elastic deformation of the speci-
successfully developed at Oxford. The specimen men.
design closely follows the proposals of Ewins When applied to composite materials several
[18], which have gained widespread acceptance major problems arise. Because of the difficulty
for the quasi-static tension testing of composite of ensuring a tensile failure in the gauge section
materials [19]. The testing technique is similar rather than a shear failure in the grips and because
to that used in the testing of isotropic materials of the need to minimize stress concentrations
[20] but incorporates an instrumented loading, or associated with the anisotropic nature of the
input, bar. Stress equilibrium across the specimen material, significantly longer test specimens are
is achieved early in the test and good agreement is required, making the attainment of stress equi-
obtained between the measurements of elastic librium across the specimen more difficult. In

1811
 ~Weighbor tube Figure 1 Schematic arrangement of tensile
Hopkinson bar. (a) Standard version, (b)
modified version for testing CFRP, and (c)
modified version for testing GFRP.

IIII1[ Ill Inertia

Jbar

Specimen

Lower
grip

addition, since for most composite materials the 3. Specimen details

total strain to failure is only a small fraction of that Unidirectionally-reinforced carbon/epoxy speci-
obtained in metal specimens, the accurate deter- mens having the dimensions shown in Fig. 3a were
mination of strain becomes very much more critical. cut with the tensile axis parallel to the direction of
For these reasons the standard tensile version
x
of the split Hopkinson's pressure bar was modified,
see Fig. lb, to include an instrumented input bar
t
preceding the specimen and inertia bar and also
sliding freely within the weighbar tube. Strain
gauges at two stations on the input bar enable
the incident and reflected waves to be monitored
in the specimen test itself. Thus the validity of
the assumption of stress equilibrium across the
Ill ---t
specimen may be directly checked and strain
determination in the specimen may be made from -B
measurements taken in a single test, eliminating
Strain
the need to compare results from two potentially gauge D -
slightly different tests. stations C -
Within the constraints of the existing weighbar
tube, however, the introduction of the input bar
limits the maximum length of the inertia bar and
hence the maximum duration of test for which the
full dynamic analysis is possible. As is apparent
from Fig. 2, which shows the Lagrange diagram for
the modified tensile Hopkinson bar, the time
interval over which the wave analysis in the input
f
bar can be carried out cannot conveniently exceed
( T 2 - T1). Since in the present apparatus /'2 ~
55 gsec and T1 = 25 #sec it is necessary that the =f
specimen should fracture within about 30/~sec. In
practice this limits the application of the technique
to uniaxially-reinforced CFRP and initial results Figure 2 Lagrange (x, t) diagram for modified tensile
will be presented for specimens of this material. Hopkinson bar.

1812
 Figure 3 Specimen design (all dimensions in
FiLLet / / mm) (a) CFRP and (b) GFRP.
rodius~ LO
F'oroL[eL
gouge section

(o) - ,41
~9.5~-

/b/ [

rodius

reinforcement from 2 mm plate supplied by Bristol ing 9 mats of Marglass 116S fine-woven fibres in
Composite Materials Ltd. The fibres were of type an Araldite epoxy matrix of MY753 resin and
HYFIL-Torayca-130-S in a proprietary resin HY951 hardener. Specimens were cut with the
system of type R7H, a modified bisphenol A tensile axis either parallel to one of the principal
medium temperature epoxy similar to Araldite reinforcing directions, 0 ~ specimens, or lying in
MY750. A typical quasi-static tensile strength of the plane of reinforcement and inclined at 45 ~ to
1.2GPa and a tensile modulus of 131GPa were both the principal reinforcing directions, 45 ~
quoted by the manufacturer for a composite specimens.
volume fraction of 60%. The parallel grip regions As an additional check on the Hopkinson bar
of the specimen were fixed into parallel-sided analysis for strain and strain rate, some tests were
slots in the loading bars using Chemlok 304 high- performed with a further set of strain gauges
strength epoxy adhesive. With a grip region of attached directly to the gauge section of the speci-
length 19ram tensile failure was obtained in the men. Techni Measure 120;2 strain gauges, type
specimen gauge region before shear failure occurred FL3A, having an active gauge region of 3 mm long
in the adhesive. Because of the relatively long by 1.gmm wide, were bonded centrally on each
grip section it was anticipated that problems face of the parallel central region of the specimen,
might arise from stress wave reflections at the using the recommended cyanoacrylate adhesive.
sections AA and BB in Fig. 2. In practice, how-
ever, the change in impedance across these sections 4. Validation of the impact testing
was so slight that any reflections resulting were technique
too small to be detected. A typical set of strain-time traces for a test in
In subsequent tests on commercially produced which additional calibration gauges were attached
(high volume fraction) unidirectionally-reinforced directly to the specimen is shown in Fig. 4a. The
glass/epoxy specimens, however, failure was small perturbation, which is regularly observed in
always found to occur by shear within the speci- such tests near the start of the transmitted strain-
men at the resin/fibre interface closest to the time trace, era, coincides in time with the break-
adhesive in the loading-bar slots. For tests on down of the specimen gauges, es, shortly before
GFRP (glass fibre reinforced plastics), therefore, the specimen fractures. It is thought to be due
a woven-roving reinforced material was used. to some resulting electrical interference. In tests
Specimens having the dimensions shown in Fig. where no gauges are attached directly to the speci-
3b were cut from 1/8 inch thick plate of "Perma- men, see Fig. 4b, no such perturbation is observed.
glass 22FE" supplied by Permali Ltd., incorporat- From the Lagrange diagram of Fig. 2 it may be
1813
Figure 4 Strain-time traces for impact tests on CFRP (Total sweep time: 100 ~sec). (a) With strain gauges attached to
specimen. Strain-time signals from: eI gauge position I; eii gauge position II; es - specimen gauges; eii I gauge
- - - - - -

position III. (b) With signals from gauge positions I and II superimposed.

seen that, in the absence of dispersion and for the stress-strain curve of Fig. 6a. The specimen
times ~< T1 identical signals should be recorded at fails after about 25#sec at a stress of 1.27GPa
stations I and II. This is confirmed in Fig. 4b and a strain, related to the gauge length of 19 ram,
where it is shown that these two signals may be i.e. from CC to DD in Fig. 2, of about 0.9%. The
superimposed almost exactly for times up to average strain rate was about 350 sec. The stress-
about 25 gsec. Standard strain gauge bridges and strain curve shows an initial linear region, corres-
two dual-channel transient recorders were used ponding to a modulus of 142 GPa, followed by a
to store the strain-time traces of Fig. 4. These slight increase in stiffness preceding failure. Also
were then subsequently displayed and photo- shown, in Fig. 6b, is the stress-strain curve for the
graphed on an oscilloscope screen while for same test but with strain measurements obtained
calculation purposes a hard copy could be pro- from the gauges attached directly to the specimen
duced on a chart recorder. and relating, therefore, to the deformation in the
Data obtained in this way was used in the
Hopkinson-bar analysis, illustrated in Fig. 5,
for a test on a CFRP specimen impacted at a 500 I (Pc) Vcc
velocity of about 10msec -1. Again the strain-
time traces from the two sets of input bar gauges
are seen to superimpose almost exactly for times
up to about 25/~sec. The subsequent difference
between these two traces is used to determine
the velocity and stress at the input end of the 300 EEl
specimen, section CC in Fig. 2. The corresponding
stress and velocity trace for the output end of the ,,7
specimen, section DD in Fig. 2, derived from the
200 [ / o 5 lo 20 13o3
inertia bar gauges, is delayed by just over 2 ~sec,
the time for an elastic wave to travel between //
sections CC and DD in a CFRP specimen. The //
stress-time traces for the two ends of the speci-
men, oee and Eein, are seen to coincide very //~j/.~" ~--EEII l (x) EEII
closely almost from the start of loading, confirm-
ing the validity of the assumption of stress equili. 0
brium across the specimen. .... . . . . . .
The corresponding strain is obtained by inte- 5 10 15 20 30 35 4.0 Z.5 50 55
grating between the velocity time curves for the Time (t~sec)
two ends of the specimen in the usual way, giving Figure 5 Hopkinson-bar analysis for impact test on CFRP.

1814
 1-6 Figure 6 Tensile stress-strain curve for impact
(o) E=142 GPo on CFRP at 10msec -1. (a) Specimen strain
& =350 sec- determined by Hopkinson-bar analysis. (b)
1'4 Specimen strain determined from strain gauges
attached to specimen.

1.2

(b) E=14z, GPo

& = 370 sec-1
1.0
g_
(_b

0.8

0.6

0-4

0.2

0It' v' I I I i I
I i i

0 0.2 0-4
S t r a i n (%)

parallel section of the specimen, i.e. for a gauge less good because of the need to take the difference
length of 3 mm. The two curves are very closely between two transient signals after adjusting for
similar. At all stress levels the specimen strain the time difference between them. The validity
gauges indicate a marginally lower strain than of the check on stress equilibrium depends both
that derived from the Hopkinson bar analysis
giving, therefore, a slightly higher initial modulus,
144GPa compared to 142GPa. As might be
expected, the specimen gauges break down
slightly before the failure of the specimen, i.e.
at 23/.tsec and at a strain of 0.86% giving an
average strain rate of about 370 sec. That a genuine
tensile failure is obtained is demonstrated in
Fig. 7 which, for a CFRP specimen impacted
at 10msec -1, shows the fracture path passing
through the strain gauges in the central parallel
region of the specimen.

5. Consideration of experimental accuracy

The close correlation between the two experi-
mental curves of Fig. 6 encourages considerable
confidence in the validity of the testing technique.
Both curves, however, rely on the same stress-
time data and on the same assumption of stress
equilibrium across the specimen. The accuracy of
stress measurement at the inertia bar gauges is
estimated to be about -+2%, after making some
allowance for dispersion and wave reflections in
the grip section, DD to BB. The accuracy of the Figure 7Tensile failure in parallel section of CFRP
stress measurement on the input side is probably specimen with strain gauge attached.
1815
on the accuracy of the stress measurement and In the test configuration of Figs. lb and 2 the
also on the ability to identify with confidence velocity at the input end of the specimen, i.e.
the time zero's of the various transient signals. Vcc at section CC, is determined from the signals
This confidence is strengthened by the use of at gauge positions I and II. If the specimen can be
transient recorders with a time resolution of treated as a thin elastic rod of known elastic con-
0.1 psec which allows an accurate correlation to stants, then an estimate of Vec may also be derived
be made between the actual wave transit times from the signal at gauge position III, using the
between the various gauge stations and those elementary theory of elastic wave propagation and
predicted by elementary elastic wave theory. calculating back through the specimen. These two
Since the specimen shows an essentially linear estimates of Vce are shown in Fig. 8 for a test at
elastic behaviour this correlation may be carried an impact velocity of 10 m sec -1 on CFRP material
through between gauge stations II and III. Allow- and are seen to be in close agreement for times up
ance for a modified elastic wave speed in the grip to about 16psec. At times greater than this the
regions, AA to CC and DD to BB, reduces the increase in stiffness of the specimen, as shown in
overall transit times by about 1.4gsec. An error Fig. 6, leads to a slight overestimate for Vcc, which
in the time measurement of this order corresponds is based on an assumed linear elastic response of
to an error of about + 3% in the strain determined the specimen. The reverse effect is observed when,
by the Hopkinson bar analysis. The stress-time at about 22psec, the specimen fractures, corres-
signals for the two ends of the specimen may also ponding to a sudden drop in stiffness. It is clear,
be displaced relative to each other, affecting the therefore, that a reasonably close estimate of the
validity of the assumed stress equilibrium and velocity on the input face of the specimen may be
modifying slightly the average stress in the speci- made from the strain-time trace on the output
men at any given time. Taking all these effects bar in an impact on CFRP material.
together the maximum error in the modulus as
determined by the Hopkinson bar analysis is 6.1. Calibration of the yoke velocity, Yy
estimated at +7GPa, i.e. about +5%. A similar In the light of this observation, an attempt was
accuracy would be expected for the results based made to calibrate for the yoke velocity in the
on strain measurements made from the speci- test configuration of Fig. lc using the near-
men gauges. linear response obtained in a CFRP specimen test
rather than by the standard calibration using a
6. Tests on GFRP material separate elastic bar. This has the important advan-
The validity of the technique described above tage that essentially the same yoke-specimen-
for the tensile impact testing of the CFRP com- inertia bar geometry is retained in both the cali-
posite has been demonstrated. Because of the high bration and the specimen tests. Strain gauges
modulus and very low strains to fracture in CFRP attached directly to the specimen were used to
specimens the accuracy of strain measurement was determine the strain in the specimen while the
the most severe problem in these tests. Accuracy specimen stress was derived from the inertia
of the strain measurement is less critical in tests on bar gauge signal, assuming stress equilibrium across
GFRP specimens where the modulus is lower and the specimen. The resulting dynamic stress-strain
the strains to fracture are higher. Other problems response, shown in Fig. 9a, was nearly identical to
arise, however, since, in general, the time to frac- that shown in Fig. 6. The initial modulus was
ture well exceeds 30/~sec and strain gauges attached 145 GPa, the fracture stress 1285 MPa, the maxi-
directly to the specimen are of only limited use mum strain 0.85% and the average strain rate
because the onset of surface damage leads to about 350sec -1. Working back from the inertia
break down of the gauges well before the final bar strain-time signal, as described above, the
fracture of the specimen. For these reasons an input, or yoke, velocity, Vy, equivalent to Vcc in
attempt was made to accurately calibrate the input the test arrangement of Fig. 2, may be estimated.
signal on the standard tensile Hopkinson bar, Fig. The velocity-time curve obtained for Vy differs
lc, using the known response of CFRP specimens, slightly from that for Vcc, as might be expected
and then to use this test configuration for GFRP since the details of the test arrangement are differ-
specimens so permitting a full dynamic analysis ent. The difference in velocity is not sufficient,
for times up to 150psec. however, to significantly affect the dynamic

1816
 Figure 8 Estimates of velocity Vee at
input end of specimen. (a) Derived from
22
inertia bar gauges, station III. (b)
Derived from input bar gauges, stations
20 I and II.

18
(o)
---"
'o 16
E

12
E

8
"5
o_
c

"6 6

>

I I I I I
0 Z, 8 12 16 2o 24 28 32
Time ( 9sec-1)

stress-strain response in a material as relatively total strain to fracture, it is necessary to use the
rate insensitive as CFRP. yoke velocity estimated from CFRP tests over the
More importantly the velocity-time curve for rising part of the velocity profile and that from the
Vy obtained in this way greatly exceeds that elastic test only in the determination of the
determined by the traditional technique, i.e. subsequent region of the stress-strain curve. In
involving a separate elastic test. Using this latter effect, the elastic part of the CFRP specimen test
technique gives the stress-strain curve of Fig. is being used as a more accurate version of the
9b for which the apparent modulus is 390 GPa. standard separate elastic test.
It should be noted that these large discrepancies
between the two estimates for Vy will only 6.2. Validation of technique for GFRP
appear during the steeply rising region of the specimens
velocity profile where small differences in the Tests were performed at an impact velocity of
experimental arrangement and in the acoustic about 15msec -1 on GFRP specimens oriented
impedance of the components involved will be of with the tensile axis parallel to one of the direc-
major significance. Subsequently the yoke velocity tions of weave (0 ~ specimens). Strain gauges
will be dependent almost entirely on the response attached directly to the gauge section of one
of the weighbar tube and so will be the same for such specimen failed at just over 1% strain and
both elastic and specimen tests. For this reason at a stress, determined from the inertia bar gauges,
little weight is given, in general, in tests on ductile of 455 MPa. Up to this point the specimen response
metal specimens to strain measurements of less had been linear, corresponding to a modulus of
than I to 2%. In tests on GFRP, however, since 44.5 GPa. This compares with a modulus of
strains of this order are a significant part of the 46 GPa obtained in the same test using the Hop-

1817
 1-6 Figure 9 Comparison of stress-
strain curves for CFRP obtained
on test arrangement of Fig. lc.
(b) E=390 GPq (a) Using strain gauges attached
1,4 ~ 160 sec-I to specimen. (b) Using separate
"elastic" test.
(o) E= 1/~5 GPo
1.2 E:~ 350 sec-1

n
t9 1.0

0-8

0.6

0-4

0.2

0-2 O-Z, 0-6 0-8 1-0 1.2

Strain (%)

kinson bar analysis to determine strain and a yoke formed on the same designs of specimen in an
velocity derived from a CFRP elastic test as des- Instron loading machine at a rate of about 10 -4
cribed above. The resulting dynamic stress-strain sec -1 and in an hydraulically-operated loading
curve is shown in Fig. 10. The close agreement machine [21] at a rate of about 10sec -1. For
between these two values of modulus encourages both machines and for each specimen material,
confidence in the validity of the technique. Some calibration tests for the elastic deflection of the
uncertainty remains, however, at higher strains, testing machine were performed using strain
beyond the fracture strain of the CFRP specimen gauges attached directly to the specimen gauge
and the failure strain of the specimen gauges in the region. Results for one such test at the inter-
GFRP test. In this region the true yoke velocity mediate rate, on a CFRP specimen, are given in
is assumed to converge towards that derived from Fig. 12. Signals showing the variation with time
the standard elastic test, see Fig. 11. Extreme of the specimen strain, es (from strain gauges
choices for the extrapolated curve in Fig. 11 attached to the specimen), the applied load P
could lead to a range in the calculated fracture (from the load cell strain gauges) and the total
strain in Fig. 10 of from 2.85 to 3.25%, i.e. a deflection, 8 of specimen and machine (from
scatter band of -+7%. In tests on 45 ~ GFRP transducers attached to the moving crosshead)
specimens the same range will apply but the were stored in a transient recorder and subse-
fracture strain is so much higher, about 11%, quently displayed on an oscilloscope screen using
that the percentage error falls to about -+2%. both the y - t and the x - y mode. In the latter an
essentially linear dependence of load both on
7. Tests at low and intermediate rates strain and on total deflection is observed.
of strain
In order to determine the effect of strain rate 8. Results
on the tensile deformation and failure of the 8.1 S t r e s s - s t r a i n response
CFRP and GFRP materials tests were also per- Stress-strain curves for CFRP specimens at three

1818
 Figure 10 Tensile stress-strain curve for 0 ~
F (a) E=z,6 GPa
& :870 seC~
GFRP specimen (impact velocity, 15 msec -1 )
(a) specimen strain determined by Hopkin-
son-bar analysis; (b) specimen strain deter-
mined from strain gauges attached to speci-
men.

0"75

0-5
e~

E= zJ..5 G P o

0.25

0-5 1.0 1.5 2.0 2.5 3-0 3-5

Strain (*/~

250 Extrapo(ated curve. .-.. ~ _ _

CFRP ~elostic specimen' test " ~ 1 1 1 " ' ~ ' ~ " ~

2oo
13-

x~ 150
u

u
2 100
>

~- 5O

0 I I I I I
0 10 20 30 a0 50
Time (p.sec)
Figure 11 Comparison of Yoke Velocities determined from standard elastic test and from CFRP elastic specimen test
(impact velocity, 10 msec-1).

1819
Figure 12 Test records at an intermediate rate of strain (CFRP specimen; strain rate about 7 sec-1). (a) Load, P, against
specimen strain, %. (b) Load, P, against total deflection, 6. (c) Specimen strain, es, against time, t. (d) Load, P, against
time, t. (e) Total deflection, 6, against time, t.

rates o f strain are presented in Fig. 13. The full ing strain rate as also does the strain at failure,
line and the quoted values of strain rate and from about 7.5% at the lowest rate to about 11.5%
modulus were derived in each case from a test in under impact loading.
which strain gauges were attached directly to the
specimen. The scatter bands refer to data from 8.2 Fracture appearance
further tests without strain gauges on the speci- For the CFRP specimens, a similar fracture mode,
men. Similar results for 0 ~ and 45 ~ GRFP speci- i.e. a tensile failure in the centre of the paralM
mens are presented in Figs. 14 and 15. A signifi- gauge region with little damage to either side of
cantly different behaviour is shown by the two the fracture surface, see Fig. 16a, was found at all
types o f material. For CFRP specimens over nearly strain rates. In contrast, a marked change in frac-
seven orders o f magnitude, no effect of strain rate ture appearance with strain rate was observed in
could be detected on either the tensile modulus, tests on 0 ~ GFRP specimens. At quasi-static
146 -+6 GPa, or the stress at fracture, 1.2 -+0.1 GPa. rates, see Fig. 16b, damage was limited to regions
In contrast, for both orientations of GFRP speci- close to the fracture surface. At increasing rates
men the maximum stress preceding failure increased the damage was found to extend further from
dramatically with strain rate, from 348 ---35 MPa at the fracture surface, covering the entire gauge
lO-4sec -1 to 8 9 9 + 2 8 M P a at 870sec -1 for 0 ~ region, i.e. well beyond the 6 mm parallel section,
specimens and from 212 + 12MPa at 2.5 x 10 -a in tests at impact rates, see Fig. 16c. In tests
sec -1 to 3 9 2 + 3 8 M P a at l l 2 0 s e c -1 for the 45 ~ on 45 ~ GFRP specimens the damage covers the
specimens. Also, in the 0 ~ specimens a marked entire gauge region at all rates of strain. In general
effect of strain rate was apparent on the modulus, the fracture surface followed steps inclined at
which increased from 19.6+0.9GPa to 48.6 + 45 ~ to the direction of loading. Differences
2.9 GPa over the same range of strain rate. For between individual specimens were observed but
these specimens the stress-strain response becomes appeared to be unrelated to the rate of straining.
increasingly nonlinear with increasing strain rate
and the strain at fracture increases from about 2% 9. Discussion
to about 3%. The 45 ~ specimens also show an 9.1. CFRP material
increase in modulus with increasing strain rate but As far as is known no other directly comparable
the effect is less marked and of the same order as data on the tensile impact behaviour of unidirec-
the accuracy of measurement. The initial linear tionally-reinforced CFRP are available. Similar
range, however, increases significantly with increas- tests, but on carbon/epoxy specimens with a

1820
 1,4

1.2

1.0

a_ 0.8
O

ul

o.6

0.4

0.2

r i i
0 0.25 0.5
Strain (%)

Figure 13 Tensile stress-strain curves for CFRP specimens (a) ~ = 5 X 10 -4 sec -1, E = 145 GPa, af = 1.21 • 0.07 GPa
(mean of 5 tests). (b) b = 7 sec -1 , E = 145 GPa, af = 1.26 • 0.07 GPa (mean of 4 tests). (c) ~ = 450 sec -~ , E = 149 GPa,
of = 1.14 • 0.05 GPa (mean of 4 tests).

plain-woven cloth reinforcement, were performed o f this difference in behaviour is not immediately
by Kawata et al. [16]. They also found the mech- apparent, nor is it clear whether it is of any real
anical response to be relatively insensitive to significance.
strain rate. Impact tests on unidirectionally-
reinforced CFRP, but in compression, were 9.2. GFRP material
performed b y Gfiffiths and Martin [9]. Although Because of the different reinforcement geometry,
they reported a dynamic modulus significantly a direct comparison with the present CFRP results
higher than the generally quoted static moduli is not possible. Tensile tests on glass/epoxy speci-
for the same class o f material specimen geometry mens with a fine-wave reinforcement have pre-
was shown to affect the shape o f their dynamic viously been performed, at intermediate loading
stress-strain curves so some doubt must remain rates b y Pink and Campbell [22] and at impact
regarding the validity o f their results. rates by Kawata et al. [16]. The strain-rate sen-
An anomaly is also apparent in the present sitivities o f the fracture stress and the fracture
CFRP results. Stress-strain curves obtained at strain obtained in these two investigations are
an impact velocity o f 10msec -1, using b o t h the compared with the present results in Figs. 17 and
modified tensile SHPB (with and without strain 18. Kawata tested 0 ~ specimens at only two strain
gauges on the specimen, see Fig. 6) and the original rates. Pink and Campbell tested b o t h 0 ~ and 45 ~
version o f tensile SHPB (also with strain gauges specimens cut from similar "Permaglass 2 2 F E "
on the specimen, see Fig. 9) show a slight increase plate (0.9 m m thick with five layers of fine-weave
in stiffness just before failure. This effect was not glass cloth) to that used in the present investigation
apparent, however, at lower rates or at an impact at some ten or more strain rates over the range
velocity o f 15msec -1, see Fig. 13, even in tests 10 -4 to 10 sec -1. Over this strain rate range their
where strain gauges attached to the specimen results show the same general trends as found here.
monitored the specimen strain. An explanation The discrepancy in absolute values of fracture

1821
 1.0[- Figure 14 Tensile s t r e s s - s t r a i n curves
for 0 ~ G F R P s p e c i m e n s . (a) ~ = 10 -4
sec -1, E = 1 9 . 6 G P a , a r e a x = 3 4 8 - + 3 5
MPa ( m e a n o f 5 tests). (b) ~ = 23
sec -1 , E = 28 GPa, a r e a x = 5 9 2 -+ 5 4 MPa
( m e a n o f 5 tests). (c) ~ = 8 7 0 sec -1 , E =
46 GPa, a r e a x =- 899 -+ 28 MPa ( m e a n o f
0-75 4 tests).

"6
0-
(c)
o
0-5
Q)

025

OV I I I i I m
l'O 2"0 3.0 4.0
Strain (~

0.4

0-3

"G
D_
(,9

ul
0.2
u)

0.1 "~-"(b)

Figure 15 T e n s i l e s t r e s s - s t r a i n curves
for 45 ~ G F R P s p e c i m e n s . (a) ~ = 2.5 X
10 -4 sec -1 , E = 11.3 GPa, a r e a x = 2 1 2 -+
1 2 M P a ( m e a n o f 5 tests). (b) ~ = 2 4
sec - t , E = 1 5 . 0 G P a , Omax = 325 -+ 26
MPa ( m e a n o f 4 tests). (c) ~ = 1 1 2 0
sec -1, E = 1 8 . 3 G P a , (rma x = 3 9 2 - + 3 8
MPa ( m e a n o f 5 tests). Strain (%)

1822
 Figure 16 Fracture appearance of composite specimens
(a) CFRP specimen after straining at a mean rate of
450 see -1 (X 8), (b) 0~ GFRP specimen after straining
at a mean rate of 10 -4sec -1 (X 8) and (c) 0 ~ GFRP
specimen after straining at a mean rate of 870 sec -1 (X
7.5).

here and in the work of Kawata et al. [16], in b o t h

investigations the same effect of strain rate on
fracture appearance was obtained. In all tests
on CFRP and in quasi-static tests on 0 ~ specimens
o f G F R P damage is confined to regions close to
the fracture plane while in impact tests on GFRP,
damage covers much o f the gauge region and the
reinforcement separates from the matrix over a
considerable distance to either side o f the fracture
surface. It is also apparent, see Fig. 14, that the
stress and strain, their results being consistently initial modulus increases with strain rate and at
lower than those reported here, may arise from the higher loads the stress-strain curve becomes
different designs o f specimen used or from differ- increasingly nonlinear. In consequence the energy
ences between the plates from which the speci- absorbed in fracturing the G F R P specimens, as
mens were cut. The results o f Kawata et al. [16] determined from the area under the stress-strain
under impact loading do not show such a dramatic curve, increases dramatically with strain rate, see
increase in fracture stress as here but indicate a Table I, whereas that for the CFRP specimens
very much higher strain to fracture. Both these remains unaffected b y the strain rate and is signifi-
discrepancies could be connected with the loading cantly less than that obtained for the G F R P
arrangement used b y Kawata et at., in that the specimens at intermediate and impact rates o f
development o f significant bending stresses on
TABLE I Energy absorbed to fracture per unit volume
impact could lead to an apparent reduction in the of specimen (MPa)
tensile load at failure while the use o f a cylindrical
Strain rate Material
specimen with a screw fixing makes the accurate
determination o f small specimen strains b y the CFRP GFRP
Hopkinson-bar analysis very difficult., 0~ specimens 45 ~ specimens
Low rate 4.6 4.1 i1.3
9.3. Fracture behaviour Intermediate
rate 5.3 10.5 19.8
Despite these discrepancies between the dynamic
fracture stresses and fracture strains determined Impact rate 5.0 18.5 32.9

1823
 1000 Figure 1 7 Effect of strain rate on maxi-

x_oo/
Present results 9 -- 0~ o --/,5 ~ / mum stress in GFRP specimens.
900 Pink & Campbell[22] f"// 0~ -'-~-- 45~ I/

?00

500 /
[3-
jJ

c)
500 _- ~ I 1 J j

2
/.00

LL

200

100F

I I I I i I I I
-Z, -3 -2 -I 0 1 2 3
Strain rate-Log(sec-I)

strain, although a direct comparison is clearly 9.4 Testing t e c h n i q u e

invalid because of the different reinforcement As discussed above, the technique developed
geometries. here for tensile impact testing of composite
This marked difference between the impact materials is either limited to specimens failing
response of carbon/epoxy composites and glass/ within about 30/~sec or requires very careful
epoxy composites has been known for some time determination of the input velocity. Although
[7] and has lead to the introduction of hybrid it has proved possible to calibrate for the input
composites where the high "toughness" of GFRP velocity, using the previously determined response
at impact rates is combined with the high stiffness of CFRP specimens, with sufficient accuracy to
of CFRP at all rates in an attempt to opimize the permit the testing of GFRP specimens to failure
overall mechanical behaviour. Most studies of the at times up to about 100/2sec, nevertheless it
impact response of hybrid composites, however, would dearly be more satisfactory to monitor
have employed Charpy type tests on notched [23] input and output stress and velocity in the same
or un-notched [24] specimens. Even though an test. The construction of an extended version of
instrumented tup may be used [5] and a load- the modified tensile SHPB to allow this for times
time history obtained, the Charpy technique up to about 150 ~sec is now in hand.
suffers serious limitations when fundamental A second limitation relates to the testing of
information on the impact response is being commercially-produced (high volume fraction)
sought so it is not surprising that conflicting unidirectionally-reinforced GFRP where, using
results have sometimes been obtained [4, 25]. The the same design of specimen as for CFRP, it has
present testing technique, however, avoids many not as yet proved possible to obtain a tensile
of these limitations and should, therefore, allow failure in the specimen gauge region. Instead
a more fundamental study to be made of the failure always occurs in shear within the specimen
impact response of composites in general and at the resin/fibre interface nearest to the slots in
hybrids in particular. the loading bars. The use of the instrumented
1824
 Figure 18 Effect of strain rate on frac-
ture strain in GFRP specimens.

12F Present resutts 0~ L5~

L 9 -- o --
11 o o /I
Pink & Compbet[ [22] 9 " / / - 0 ;\\"~- L5 /I
[ I

10 Kawata etat.[16] x - 0~ / ..L/x/

9
//"
8

~ 6 /
0A

~s
L~

jx
3

1 I ! I I I
-~ -3 -2 -1 0 1 2
Strain rate-tog (sec-1)

input bar in the extended version of the present are found to be independent of strain rate. In
impact tester will be required to monitor the contrast both orientations of GFRP specimen
effects of any changes in specimen design neces- showed a dramatic increase in failure strength at
sary in the solution of this problem. impact rates of strain, a significant increase in
failure strain and, for the 0 ~ specimens, a marked
10. Conclusions increase in initial modulus. The change in mech-
A modified version of the tensile SHPB apparatus anical response with increasing strain rate was
has been successfully developed for the tensile associated, in the 0 ~ specimens, with a change in
impact testing of unidirectionally-reinforced CFRP. the fracture appearance, limited matrix cracking
Stress equilibrium across the specimen is attained close to the fracture surface at low rates extend-
at an early stage in the test. Specimen strain is ing to cover the entire gauge section at impact
determined to an accuracy of about -+3% and the rates where extensive debonding between the
tensile modulus to about -+5%. The technique has fibres and the matrix was also observed.
been successfully extended to the testing of
GFRP specimens having a woven-roving reinforce- References
ment with the tensile axes paralM to, or at 45 ~ to, 1. G. DOREY, "Fracture behaviour of carbon fibre
the principal reinforcement directions. composites su~ected to impact loads", AGARD
Over a range from about 10 -4 sec -1 to about Conference Proceedings 163 on Failure modes
of composite materials with organic matrices and
1000sec -1 the modulus, fracture strength and their consequences in design. AGARD-CP-163,
failure mode of unidirectionally-reinforced CFRP 1974.
1825
 2. A.E. ARMI~NAKAS and C.A. SCIAMMARELLA, 14. T. FUJII and M. MIKI, in Proceedings of the Sym-
Exp. Mech. 13 (1973) 433. posium on Mechanical Behaviour of Materials,
3. A. ROTEM and J. M. LIFSHITZ, in Proceedings of Kyoto, Japan, 1973, p. 83.
the 26th Annual Technical Conference, SPI Reinfor- 15. I.M. DANIEL, R.H. LA BEDZ and T. LIBER,
ced Plastics/Composites Division (Society of Plastics Exper. Mech. 21 (1981) 71.
Industry, New York, 1971) Paper 10-G. 16. K. KAWATA, A. HONDO, S. HASHIMOTO, N.
4. G. MAROM, S. FISCHER, F. R. TULER and H. D. TAKEDA and H. L. CHUNG, in Proceedings of the
WAGNER, J. Mater. Sci. 13 (1978) 1419. Japan - US Conference on Composite Materials,
5. D.F. ADAMS and A. K. MILLER, ibid. 11 (1976) Tokyo, January 1981, edited by K. Kawata and
1697. T. Akasaka (Japan Society for Composite Materials,
6. M.G. PHILLIPS, "Fracture and fatigue of hybrid Tokyo, 1981) p. 2.
composites", in Fibre Composite Hybrid Materials, 17. J. HARDING and L.M.WELSH, presented at the
edited by N. L. Hancox (Applied Science Publishers, 4th International Conference on Composite Materials,
Barking, Essex, 1981) Chap. 4. Tokyo, October 1982 (Japan Society for Composite
7. D.F. ADAMS, ASTM STP 617 (1977) p. 409. Materials).
8. R.L. SIERAKOWSKI, G. E. NEVILL, C.A. ROSS 18. P.D. EWINS, RAE Technical Report No. 71217
and E. R. JONES, d. Comp. Mater. 5 (1971) 362. (1971).
9. L.J. GRIFFITHS and D.J. MARTIN, or. Phys. D. 19. "Grafil test methods", Courtaulds Ltd., Carbon
AppL Phys. 7 (1974) 2329. Fibres Unit, PO Box 16, Coventry, 1977.
10. T. PARRY and J. HARDING, Colloque International 20. J. HARDING, E.O. WOOD and J. D. CAMPBELL,
du CNRS No. 139, Plastic behaviour of anisotropic J. Mech. Eng. Sci. 2 (1960) 88.
solids, Grenoble, June 1981 (OUEL Report No. 21. R.H. COOPER and J.D. CAMPBELL, ibid. 9
1365/81). (1967) 278.
11. J. HARDING, in Proceedings of the Conference on 22. E. PINK and J.D. CAMPBELL, J. Mater. ScL 9
Mechanical Properties at High Rates of Strain, (1974) 658.
Oxford, April 1979, edited by J. Harding, Conf. Ser. 23. D.F. ADAMS, ibid. 10 (1975) 1591.
No. 47 (Institute of Physics, Bristol and London, 24. R.R. DIXON, 3-5th SPE Conference, USA (1977)
1979) p. 318. p. 344.
12. E. McABEE and M.CHMURA, in Proceedings of 25. B. HARRIS and A . R . BUNSELL, Composites 6
the 16th Annual Technical Conference, SPI Rein- (1975) 197.
forced Plastics Division, Chicago (Society of Plastics
Industry, New York, 1961) Paper 13-D.
13. B. REVSIN and S.R. BODNER, Israel J. Tech. Received I October
7 (1969) 485. and accepted 23 N o v e m b e r 1982

1826