Composites Part B 171 (2019) 310–319

Contents lists available at ScienceDirect

Composites Part B
journal homepage: www.elsevier.com/locate/compositesb

A Split-Hopkinson Tension Bar study on the dynamic strength of

basalt-fibre composites
G.C. Ganzenmüller a, b, *, D. Plappert b, A. Trippel b, S. Hiermaier a, b
a
Fraunhofer Ernst-Mach-Institute for High-Speed Dynamics, EMI, Freiburg i. Br, Germany
b
Albert-Ludwigs Universit€
at Freiburg, Institute for Sustainable Systems Engineering, Freiburg i. Br, Germany

A R T I C L E I N F O A B S T R A C T

Keywords: This paper investigates the strain rate sensitivity of laminated composites made of plies of unidirectional basalt
Mechanical testing fibres and epoxy resin. We consider laminates with quasi-isotropic [0∘, 45∘,þ45∘,90∘]s and orthogonal
Strength [þ45∘, 45∘]4 layup. A Split-Hopkinson Tension Bar is used to generate accurate stress/strain data at elevated
Basalt fibre
rates of strain of �3 � 102/s. Moderate strain rate effects are observed with strength increase of �3.5% per
decade of increased loading rate for both laminate types.

1. Introduction only exhibit strength increases of approximately 1%–10% per order of

The purpose of this work is to present reasonably accurate dynamic
tensile strength data for composites made of plies of endless, unidirec­
tional basalt fibres and epoxy resin. Over the last two decades, basalt
fibres have come into consideration as potential reinforcement for
composite materials requiring high strength and temperature resistance.
In contrast to the infamous asbestos fibre, which meets these application
requirements from a mechanical point of view, basalt is not considered
carcinogenic [1]. It is a mineral belonging to the group of silicates and as
such similar to glass in its chemical composition but with more iron and
less calcium content. While glass fibres are completely amorphous
solids, basalt fibres can feature some degree of cristallinity [2]. Basalt’s
mechanical properties strength and stiffness are lower than those of
carbon fibres, and more akin to those of S-2 glass fibres. The quasi-static
properties of basalt fibres and its composites are already well known [3,
4]. A promising field of application for basalt fibre composites are
structural elements in the automotive sector [5], where it is important to
predict the dynamic material behaviour at elevated strain rates relevant
at crash, which are on the order of 10–100/s [6]. However, as of now,
only a few studies are available on dynamic properties. This is particu­
larly true for composites composed of plies of endless, unidirectional
(UD) fibres. To our knowledge, only one, very recent study is available
[7], where the authors report that the dynamic tensile strength of a pure
0∘ basalt/epoxy composite doubles its strength when strain rate is
increased from 10/s to 300/s. This amount of strain rate sensitivity
appears exceptionally high, as UD composites of carbon or glass fibres
magnitude of increased strain rate [8–14]. In another study, the strain
rate sensitivity of woven basalt/epoxy composites was studied, and a
similarly exceptional high strain rate sensitivity was observed [15]. Is
there something special about basalt fibres which makes them very
strain rate sensitive? To shed some light on this issue, we investigate the
strain rate dependency of the tensile failure strength and strain for
quasi-isotropic [0∘, 45∘,þ45∘,90∘]s and orthogonal [þ45∘, 45∘]4 layups
of unidirectional basalt fibres in a thermoset epoxy matrix within the
strain rate regime 10 3 – 3 � 102/s. To obtain accurate data at high rates
of strain, we employ a Split-Hopkinson Tension Bar, for reasons we feel
important to briefly review in the following:
Acquiring accurate stress/strain data at high rates of strain >100/s is
not a trivial task, as has been discussed extensively in the literature [16,
17]. The difficulty may be attributed to three issues: (i) The testing
apparatus must achieve a well defined, e.g., constant velocity at which
the sample is loaded, typically on the order of a few m/s. This requires
strong acceleration and good velocity control of the gripping devices
used to interface the specimen with the testing apparatus. (ii) The stress
state within the specimen needs to be inferred from a force sensor, which
is usually not mounted on the specimen itself but rather located in-line
with the loading axis at some distance from the specimen. During the
short time span of a dynamic experiment, propagation effects of elastic
waves must be considered. The force sensor will, in general, not sense
the force experienced by the specimen at a given instant of time, but
instead report another value which is influenced by its own acceleration.
Furthermore, the dynamic bandwidth, i.e., its frequency response

* Corresponding author. Fraunhofer Ernst-Mach-Institute for High-Speed Dynamics, EMI, Freiburg i. Br, Germany.
E-mail address: georg.ganzenmueller@inatech.uni-freiburg.de (G.C. Ganzenmüller).

https://doi.org/10.1016/j.compositesb.2019.04.031
Received 22 January 2019; Received in revised form 2 April 2019; Accepted 26 April 2019
Available online 4 May 2019
1359-8368/© 2019 Elsevier Ltd. All rights reserved.
G.C. Ganzenmüller et al.
function, must be adequate to accurately describe the specimen’s force
signal. Typically, dynamic bandwidths on the order of 1 MHz are
required for experiments at strain rates between 102 – 103/s [18]. (iii)
Determination of the specimen strain state requires fast and robust
measurement techniques. Fast, because for sampling the strain evolution
of a specimen with failure strain of 5% at a strain rate of 100/s with only
100 data points, already an acquisition rate of 200 kHz is required.
Robust, because the sampling method must yield reliable measurements
for a rapidly accelerating and deforming specimen. While strain gauges
meet the requirement of fast acquisition rates, it can be difficult to
maintain adhesion on the surface of a specimen until failure. Alterna­
tively, high-speed imaging and optical strain analysis is a universal yet
expensive solution if high spatial resolution is required.
For conducting experiments at high rates of strain, two popular
experimental methods are available. The first method is given by the
universal testing machine, powered by a servo-hydraulic actuator. While
these machines can operate over large ranges of testing velocities and
forces with typical upper limits of 10 m/s and 250 kN, difficulties arise
for accurately measuring the force signal of a specimen under rapid
loading. Taking reference to the points (i) and (ii) from above, this is due
to the following reasons: (I) The apparatus needs to achieve a high
testing velocity before the specimen is loaded. This is realized via a lost
motion device, which lets the hydraulic piston travel freely under ac­
celeration. Once the piston is at speed, its movement is coupled
instantaneously to the specimen’s grips. The instantaneous acceleration
of the – typically heavy – specimen grips causes a strong inertial force
response which overlays the true force signal of the specimen’s stress
response. This effect may be theoretically visualized by considering the
frequency response of a velocity signal with a step-like acceleration. Its
Fourier transform shows amplitudes over an infinite frequency spec­
trum. (II) Force measurement is conventionally performed via a piezo-
electric load cell with finite stiffness and its own limited frequency
response. Such force sensors act as low-pass filters. The fact that the
quasi-infinite frequency spectrum associated with the step loading
character of the slack adapter is not present in the output signal is caused
by this filtering effect. While servo-hydraulic universal testing machines
have their warranted uses for specific applications, acquiring high-
quality stress-strain data, in particular for lightweight, non-metallic
materials, is challenging. Naively applying such a machine at high
rates of strain, one does not measure the true specimen response but
instead a convolution of the specimen response with the acceleration
response of the specimen grips and elastic wave reflection effects with
visible oscillatory artifacts, see e.g. Ref. [19]. However, it needs to be
pointed out that high-quality data may also be obtained using this
approach if its limitations are considered carefully [17].
A more straightforward alternative for high rate testing is given by
the Split-Hopkinson Bar method, also known as the Kolsky Bar method.
It is out of the scope of the present work to detail this technique
extensively, instead the reader is referred to Ref. [18]. Here, the spec­
imen is sandwiched between two long and straight bars of constant cross
section, the input and output bars. An elastic wave of well defined
amplitude is created in the input bar, either by impacting the bar with
another long bar, the striker, or by pre-tensioning the bar and employing
a release mechanism. The elastic wave then passes through the specimen
into the output bar, creating secondary transmitted and reflected waves
at the interfaces between specimen and input/output bars. These waves
are measured using strain gauges mounted on the bars, and analysed
using the theory of one-dimensional elastic wave propagation to yield
both specimen strain and stress. Due to the simple geometry, all 1D wave
effects are accounted for, and the resulting stress/strain curves can be of
high quality. However, this fortunate situation applies only to
compression experiments performed with the Split-Hopkinson Pressure
Bar (SHPB). Here, a small amount of grease is sufficient to stick a cy­
lindrical sample between the faces of input- and output bar, and no
explicit specimen grips are required. Due to this simplicity, the SHPB
method is widely accepted for reliably producing accurate compression
311
Composites Part B 171 (2019) 310–319
stress/strain data for composites, see e.g. Refs. [20,21], and a strong
theoretical basis is available which rationalizes these beliefs [22].
In contrast, Split-Hopkinson Tension (SHTB) Bar experiments are
more difficult to perform. The specimen needs to be attached to the ends
of the input and output bars without introducing significant additional
mass, e.g., specimen grips as used in conventional testing machines. The
presence of such additional mass means additional changes in acoustic
impedance, causing unwanted wave reflections. This violates the as­
sumptions made in the classic analysis based on simple wave propaga­
tion in media with constant acoustic impedance. The situation is
fortunate for specimens with cylindrical shape, as these can be threaded
at the ends and screwed into matching interior threads in the bar ends.
Thus, no additional mass is introduced for affixing the specimen. Com­
posites, however, are typically tested in a planar strip form. Until now,
the only solution has been has been to introduce slots either into the bar
ends or into adapters, and bond the specimens into these slots with high-
strength adhesives [9,12,23]. This process requires intensive labour and
expensive machining. Recently, two of the current authors introduced a
clamping device which allows reversible mounting of strip-type speci­
mens without violating the requirement of constant acoustic impedance
[24]. Here, we utilize this clamping device to obtain SHTB data of
reasonable accuracy and compare this data with results from quasi-static
and moderate strain rates.
The remainder of this article is organized as follows: The composite
investigated here and the experimental methods employed are described
in the next section. We then report our findings for strength and failure
strains, analyse the strain rate dependency and conclude with a critical
review of our data in the light of other researchers’ findings.
2. Materials and methods
2.1. Material specimens
Sheets of cured UD basalt fibre composites with layups
[0∘, 45∘,þ45∘,90∘]s and orthogonal [þ45∘, 45∘]4 were provided tested.
This composite is made from a prepreg of basalt fibres with filament
diameter 17 μm and a low-viscosity epoxy resin system. Laminates were
cured in an autoclave process resulting in fibre volume fraction of 60%.
To illustrate the mechanical properties of this material, we quote the
nominal strength and stiffness of pure 0∘ specimens according to ISO-527
as 1310 MPa and 44 GPa [25].
Testing specimens were cut from the sheets using a diamond coated
saw blade to obtain good cut surface quality. The dimensions of the
quasi-isotropic (QI) specimens are length � width � height ¼
100 � 3 � 1.6 mm3 with a nominal gauge length of 20 mm. The di­
mensions of the �45∘ specimens are 100 � 5 � 2.0 mm3 with the same
nominal gauge length. The same specimen geometry was used across all
strain rates. Specimens were bonded to grooved 7075 aluminium tabs
with a high-strength, thermally activated epoxy resin according to
Fig. 1. We use split tabs as these can be easily manufactured in large
quantities using a CNC router. Additionally, the split tabs provide good
accessibility to the interior which makes it trivial to correctly apply the
adhesive.
The tabs feature holes for M4 through-bolts to affix the assembled
specimen to the aluminium grips as shown in Fig. 2. The clamps are
designed to fulfill three task: (i) they provide an ISO M12 � 1.5 thread to
interface with either the Split-Hopkinson bars or the universal testing
machine. (ii) The transition from the circular interfacing end with the
testing apparatus to the rectangular shape used to attach the specimen
maintains a constant cross section area. For dynamic testing, the Split-
Hopkinson bars are also made of aluminium, such that constant acous­
tic impedance is obtained. This avoids unwanted wave reflections due to
the specimen gripping device, see Ref. [24] for details. (iii) The ø4.1 mm
through holes allow for clamping the specimen inside the grip. To this
end, lightweight M4 aluminium bolts with strength class ISO 5.6 may be
used, yielding an axial force of 1.8 kN per bolt. With 4 bolts and a
G.C. Ganzenmüller et al. Composites Part B 171 (2019) 310–319

Fig. 1. CAD rendering of the specimen with its split aluminium tabs. Also shown is the fixture which is used to align all parts during assembly and curing of the epoxy
adhesive. All dimensions in mm.

Fig. 2. Clamping fixture maintains cross section area and thus provides constant acoustic impedance which is required for reflection-free wave transmission. All
dimensions in mm.

312
G.C. Ganzenmüller et al.
conservatively assumed friction coefficient μ ¼ 1.0 between the clean
and unlubed aluminium surfaces, this translates into a useful upper force
limit of �14 kN for this clamp. This limit exceeds the expected forces in
this work by a factor of 4. We note that those parts of the through-bolts
and nuts protruding from the surfaces introduce additional mass, which
causes unwanted wave reflections. However, this effect is so minor that
its effects cannot be observed in our data.
2.2. Quasi-static and low strain rate testing
For the quasi-static and low strain rates of 10 3/s and 10 1/s, we use
a screw-driven universal testing machine with 100 kN capacity and a
maximum testing velocity of 1500 mm/min. Specimens were attached
using the gripping device described in section 2.1, which itself was
mounted to the testing machine using cardanic joints to prevent off-axis
loading. The testing velocity was adjusted in pilot experiments such that
the target strain rates were obtained. Force was measured via the ma­
chine’s internal load sensor, which is calibrated to accuracy class 1/ISO
7500–1. Strain was measured optically using a black/white camera with
resolution of 1280 � 1024 pixels and a maximum frame rate of 1500 fps
at this resolution. To this end, white contrast marks were painted on the
specimen at the edges on the gauge section with a paint marker. The
displacement of the contrast marks was recorded and converted into a
strain time series using a motion analysis software based on Digital
Image Correlation (DIC). Both force and strain time series were com­
bined into nominal stress/strain graphs by eliminating the time infor­
mation. We note that for this universal testing machine in combination
with the tested specimens, a strain rate of 10 1/s is the most dynamic
experiment that can be achieved before acceleration effects cause
ringing artifacts in the force signal can be observed.
2.3. Dynamic testing apparatus
The Split-Hopkinson Tension Bar (SHTB) used here is sketched in
Fig. 3 and described in detail in Ref. [26]. Compared to other SHTBs, this
setup is optimized for low velocities, low forces and a long pulse dura­
tion of 1.2 ms. Specimens were attached to the bars using the grips
detailed in section 2.1. Shortly before performing the experiment, white
contrast marks were painted on the specimen at the edges of the gauge
section using a paint marker. We have observed that it is crucial for the
paint to be still compliant. If it is fully dried it will come off the specimen
before actual failure occurs, thus voiding the strain measurement. For
determining the strain, we employ a line-scan camera to track the
displacement of the contrast marks. If only uniaxial displacement is of
interest, line-scan cameras are advantageous over area cameras due to
higher 1D resolution and increased light sensitivity due to larger pixel
size. We employ a model with 1 � 4096 pixel resolution and a line scan
frequency of 200 kHz. The strain is ultimately obtained by
post-processing the line-scan data using a pattern matching algorithm
with sub-pixel accuracy [27]. Force is measured via a pair of conven­
tional strain gauges mounted on the output bar, connected diagonally in
Figure 3. Sketch of the SHTB setup employed in this work. All dimensions in mm. Input and output bars are 16 mm diameter aluminium rods. The striker is a hollow
aluminium tube of 40 mm outer diameter and 20 mm inner diameter. Two strain gauge stations on the input bar and output bar measure the incident wave, εinc, and
transmitted wave, εtra.
313
Composites Part B 171 (2019) 310–319
a Wheatstone bridge circuit to eliminate bending information. The
Wheatstone bridge circuit is driven in constant voltage mode and its
output is increased by a factor of 100 using an amplifier with 1 MHz
bandwidth. This signal is recorded by a data acquisition (DAQ) card
operating at 10 MHz and 16 bit resolution. The conversion factor from
strain to force is established via a calibration procedure, wherein a
dedicated force sensor is placed between input- and output bars, while a
static load is applied onto the bars. The signal of the force sensor is
recorded using the same amplifier and data acquisition card as during a
real experiment. The force sensor, in turn is calibrated against the
already calibrated load cell in the universal testing machine described
above. Thus, the entire force measuring system consisting of strain
gauges, amplifier and DAQ card is calibrated, which compensates for
any eventual misalignment of the strain gauges or similar constant
systematic errors.
In addition to the line-scan camera, we also use a high-speed area
camera to obtain images of the deformation process. The highest sam­
pling rate of this camera is 38.65 kHz at a resolution of 336 � 96 pixels.
This is not enough for accurate strain determination, but serves well to
obtain a qualitative impression of the failure behaviour, see Appendix A.
Both line-scan and area cameras are triggered by the DAQ card to
start recording when the incident wave created by the striker reaches
strain gauge no. 1, c.f. Fig. 3. In the case of the line-scan camera, a
synchronization signal is fed back to the DAQ card, so precise timing
information is available to correlate strain and force signals. It is
important to realize that even though strain and force signals are
perfectly well synchronized in this manner, these signals are measured
in different locations: strain is measured on the specimen but force is
inferred from the strain gages on the output bar, in our case 200 mm
upstream in the transmitted elastic wave direction. The time taken by
the elastic wave to travel this distance must be accounted for by shifting
the force signal by Δt ¼ 200 mm/5090 mm/ms ¼ 0.039 ms, where
5090 mm/ms is the longitudinal wave speed in aluminium.
In classic Split-Hopkinson Bar experiments, a check of dynamic
equilibrium is possible by placing strain gages symmetrically around the
specimen, i.e., also on the input bar near the specimen. This allows to
check when the forces on both sides of the specimen are equal in
magnitude, which marks the beginning of measured data validity and
constant strain rate. Here, we cannot provide this information as no
additional strain gauge was mounted on the input bar. However, the
onset of dynamic equilibrium may also be calculated analytically, see e.
g. Ref. [22]. Usually, the time taken for the elastic wave to perform 4 to 5
round-trips within the specimen is sufficient [28,29]. In case of the
quasi-isotropic specimen, the longitudinal wave speed
is � 3300 mm/ms, and the gauge length is 20 mm, resulting in an
equilibration time of <60 μs.
Our SHTB experiments were set up such that the incident wave at­
tains a force amplitude of 20 kN, corresponding to a particle velocity of
7.2 m/s. This marks the upper end of what is safely attainable with our
aluminium setup due to its mechanical strength. Fig. 4 shows the
recording of such an experiment on a quasi-isotropic specimen. The
G.C. Ganzenmüller et al.
Figure 4. Bottom plot: Recordings from a Split-Hopkinson Tension Bar
experiment on a quasi-isotropic specimen. Both incident and transmitted waves
are direct recordings from the strain gauges at the locations described in Fig. 3,
scaled to units of force. The reflected wave is computed from the expression
εref ¼ εinc εtra . Top plot: strain rate as obtained from the reflected wave and its
time integral, which is the specimen strain.
incident wave has a pronounced rise time of �0.2 ms, which is due to the
use of a pulse shaper between striker and transfer flange, a rubber disk of
1 mm thickness. This pulse shaper significantly reduces oscillations in
the transmitted force signal, which would otherwise completely domi­
nate. However, the pulse shaper also implies that the strain rate is not
constant at the start of loading but instead slowly ramps up. In this case,
the pulse shape is chosen such that a nearly constant strain rate is ob­
tained before failure. The strain rate may be obtained in two different
ways, either directly from the strain gauge signals or by measuring
locally on the specimen using an independent method. In the following,
we will compare these two approaches and argue why it is important for
the SHTB to only use the independent local measurement.
As a result of one-dimensional wave propagation theory, the classi­
cally used strain rate is proportional to the reflected wave originating at
the interface between input bar and specimen [18],
2 c0 εref
ε_ classic ¼ ; (1)
L0
where c0 is the wave speed in the bar and L0 is the initial separation of
input and output bar, which is equal to the specimen length. This strain
rate thus measures the relative movement of the specimen-bar in­
terfaces. With our setup, we cannot measure the reflected wave εref
directly, as the geometry does not allow an isolated reading of the re­
flected pulse because the striker has a larger generalized impedance than
the input bar which causes superpositions of waves. Instead, we
approximate the reflected wave by assuming the condition of dynamic
equilibrium, also referred to as the two-wave approximation:
εref ¼ εinc εtra (2)
The strain rate computed in this fashion, along with the resulting
strain is depicted in the upper part of Fig. 4. This strain rate attains a
value of ε_classic�600/s near failure and the failure strain is 7.6%. It is
known that this approach works well in the case of a pressure bar, but we
claim that is not apt in the case of the SHTB, where the specimen
mounting introduces additional compliance. In the present case, the
specimen is glued into a holder, and the deformation of the adhesive
contributes erroneously to the nominal strain measured using the re­
flected wave. It is therefore necessary to employ a local measurement of
specimen strain.
314
Composites Part B 171 (2019) 310–319
Figure 5. Force and strain time series for a single Split-Hopkinson Tension Bar
experiment on a quasi-isotropic specimen at strain rate 300/s. Force data (black
solid line) is sampled at 10 MHz, local specimen strain (red circular symbols) is
sampled at 200 kHz using a line scan camera. The solid red line is a linear
regression to the local strain data points with slope 0.295/ms, corresponding to
the target strain rate of � 300/s. The classic, but erroneous, Split-Hopkinson
estimate of the specimen strain as computed from the reflected wave is
also shown.
Fig. 5 compares the local specimen strain measured using the line
scan camera with the strain computed from the reflected wave. The
locally measured strain is 4.9% and thus approximately only 65% of the
classic estimate. Additionally the local strain rate as obtained from a
linear fit to the specimen strain is 300 � 20/s, which is half of what is
estimated by the classic approach. We would like to emphasize that this
difference is not due to the two-wave approximation, but instead caused
by the compliance of the specimen holding fixture.
Our SHTB setup with its long pulse duration allows to reach rela­
tively low strain rates compared to what is common in the SHTB com­
munity. In principle, we could have realized even lower strain rates for
same-sized specimens: in the quasi-isotropic case, only 0.25 ms of the
available pulse duration of 1.2 ms is used, and for �45∘ specimens,
failure is reached at approx 0.55 ms. However, such experiments would
deliver little additional insight as strain rate effects typically depend
logarithmically on the strain rate. It was therefore our aim to achieve the
highest possible strain rate. The quality of our data compares well with
other Split-Hopkinson studies [8,9,12,23], and we feel confident that
quantitative conclusions can be drawn from these data.
3. Results and discussion
3.1. Data reduction
Fig. 6 shows the stress/strain curves obtained at each combination of
strain rate and laminate layup. Each testing series consists of N ¼ 5 valid
experiments. An experiment is considered valid, if failure occurred
within the gauge section, i.e., between the holding tabs. Within one
series, the individual experiments exhibit scatter which necessitates an
averaging procedure before multiple testing series can be compared to
each other. The characteristic properties maximum stress, σ max and
strain at maximum stress, εmax are represented using straightforward
averages and uncertainty estimates, e.g. for the stress: σmax ¼
PN qffiP
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
N
j¼1 σ max;j =N and sðσ max Þ ¼ j¼1 ðσ max;j σmax Þ2 =ðN 1Þ.
In contrast, computing an average stress/strain curve for an entire
testing series is not as straightforward. Within each testing series, the
sequences of strain values εi are not the same for all experiments; they
G.C. Ganzenmüller et al. Composites Part B 171 (2019) 310–319

Figure 6. Stress/strain curves for quasi-isotropic and �45∘ specimens at different strain rates. Black lines show individual experiments and red lines represent
average behaviour obtained with the histogram approach (see text). Error bars indicate the standard deviation of the average. An artificial steep drop in the average
curves indicates our definition of failure strain, which is the strain at maximum stress.

Fig. 7. Comparison of average stress/strain data for quasi-isotropic laminates Fig. 8. Comparison of average stress/strain data for �45∘ laminates tested at
tested at different strain rates. Errorbars indicate the standard deviation and are different strain rates. Errorbars indicate the standard deviation and are only
only shown for the highest strain rate results to avoid clutter. shown for the highest strain rate results to avoid clutter.

span different ranges and are sampled at different locations. Thus, a

naïve averaging process over j ¼ 1.N datasets, such as σðεi Þ ¼
PN
j¼1 σ j ðεi Þ=N, cannot work. Instead, we employ a histogram-based
Table 1
approach. First, a strain axis which extends from the smallest observed Average stress and strain values for quasi-isotropic and �45∘ laminates, for the
strain value (typically 0) to the average strain at maximum stress is different strain rates considered in this work. The discrete standard deviation is
defined and discretized with sufficiently high resolution – 1000 points in used as uncertainty estimate following the � symbol. Note that for the �45∘
our case. Each individual stress/strain dataset is then interpolated using laminate, maximum strain is defined as the strain at maximum stress.
linear functions and sampled at every histogram bin which is covered by strain rate 10 3/s 10 1/s 3 � 102/s
the dataset. Now, each histogram bin contains between 1 and N samples,
QI max. strain [%] 3.3 � 0.4 3.9 � 0.2 4.8 � 0.1
which allows the definition of an average and an uncertainty estimate, QI max. stress [MPa] 428 � 43 452 � 42 524 � 38
for which we use the discrete standard deviation. �45∘ max. strain [%] 15.0 � 1.4 14.0 � 1.2 13.9 � 0.8
�45∘ max. stress [MPa] 179 � 6 194 � 7 217 � 7

315
G.C. Ganzenmüller et al.
Fig. 9. Strain rate sensitivity analysis of failure stress for quasi-isotropic (solid
circles) and �45∘ specimens (crosses). Error bars denote one standard deviation.
Straight lines are fits using a sensitivity model which is proportional to the
logarithm of the strain rate, see text.
Fig. 10. Photographs of representative specimens after testing. The upper part shows the �45∘ specimens with zoomed-in regions for the two different strain rates
10 3/s and 300/s. The lower part shows the same for quasi-isotropic specimens. The edge length of a grey square on the background paper is 10 mm.
316
Composites Part B 171 (2019) 310–319
3.2. Quasi-isotropic specimens
Stress/strain curves for the quasi-isotropic specimens are shown in
Fig. 6, along with their representative average curves. We note that the
results obtained with the universal testing machine show some un­
dulations at strains <1%, which is due to initial movement of the car­
danic joints. Nevertheless, we are confident that the average curves are
reliable. The material exhibits almost linear-elastic, brittle character.
Maximum stress and strain at maximum stress coincide with the failure
stress and strain. Comparison among the different strain rates shows that
the failure stress and increases with strain rate from approximately 428
to 524 MPa, i.e. by 22 � 8%, see Table 1. To illustrate that the
comparatively small specimen size used here allows representative re­
sults to be obtained, we note that the corresponding quasi-static strength
according to ISO-527 using much larger, 25 mm wide specimens is
436 � 10 MPa [25]. Failure strain increases significantly from approxi­
mately 3.3%–4.8%. Classic visco-elastic behaviour, i.e., increased stiff­
ness and decreased failure strain with strain rate, behaviour is not
observed. Instead, and according to Fig. 7, all curves are superimposed
but extend to higher stress and strain as the strain rate is increased.
Within the uncertainty estimate, no distinction in stiffness can be
observed, see Fig. 10. The location at which failure occurred exhibited
little systematic behaviour, independent of strain rate. Experiments
G.C. Ganzenmüller et al.
were only considered valid, if failure occurred in the gauge region. The
failure mode could not be identified as the specimens disintegrated
instantaneously at failure due to elastic energy release. This process was
too fast to be captured even with the high-speed camera at a frame rate
of 38.65 kHz, see Fig. A11.
3.3. �45∘ specimens
The �45∘ specimens exhibit classic strain rate dependent pseudo-
viscoplastic behaviour, see Fig. 8 Following an initial linear response,
which is the same at all strain rates within the uncertainty estimate, a
rounded transition to an inelastic regime with constant strain hardening
is observed. According to Table 1 the stress level of the inelastic regime
increases significantly with strain rate: the maximum stress increases
from approximately 179 to 217 MPa, i.e. by 21 � 4%. The corresponding
quasi-static results for conventionally-sized specimens of 25 mm width
according to DIN ISO-527 is 203 � 8 MPa [25]. This differs slightly but
finite-size effects are expected especially for this specimen type as the
amount of shear deformation that can be sustained is proportional to the
specimen width. Failure strain (strain at maximum stress) is approxi­
mately 14% and thus much higher than in the quasi-isotropic case. The
failure strain appears to decrease slightly with increased strain rate,
although this observation cannot be fully justified given the uncertainty
estimate. No qualitative distinction could be made in failure behaviour
at different strain rates. Failure occurred typically in the center of the
gauge region between the holding tabs, and was less explosive in char­
acter compared to the quasi-isotropic specimens, see Fig. 10.
3.4. Analysis of strain-rate effects
We analyse the strain rate dependency by plotting failure stress
against the decadic logarithm of strain rate, see Fig. 9. Both quasi-
isotropic and �45∘ datasets are well described with a simple model,
which is proportional to the logarithm of the strain rate, i.e.,
� � ��
ε_
σ fail ¼ A � 1 þ B log : hinder wave propagation. Our results show that the strain rate sensi­
ε_ 0
Here, ε_0 ¼ 1/s is a reference strain rate which serves to render the
argument of the log function dimensionless; A is the failure stress at the
reference strain rate and B is the slope, i.e., the strain rate sensitivity.
Fitting this function to the data points from Table 1 using the Levenberg-
Marquardt algorithm yields the parameters A and B including their
uncertainty estimates. As is obvious from Fig. 9, the failure stress at ε_0 is
higher for the quasi-isotropic specimens than for the �45∘ specimens,
A ¼ 477.2 � 5.0 MPa vs. A ¼ 200.1 � 0.5 MPa. More interestingly, the
strain rate sensitivity is identical within the uncertainty estimate: We
observe B ¼ 0.038 � 0.005 in the quasi-isotropic case and
B ¼ 0.035 � 0.001 for the �45∘ specimens. This is somewhat surprising,
as most of the load within the quasi-isotropic specimen is carried by the
fibres of the 0∘ ply, and the strain rate sensitivity of the fibres itself is
believed to be weak [30]. In contrast, in the �45∘ specimens, the epoxy
matrix is loaded in shear, and it is well known that epoxy resins exhibit
pronounced strain rate sensitivity [31,32]. However, these simple
317
Composites Part B 171 (2019) 310–319
arguments may be misleading, as the composite’s strength is also a
function of the fibre-matrix interface strength. To the knowledge of the
authors, no published information is available about the loading rate
sensitivity of interface properties in the case of basalt fibres and epoxy
resins.
4. Discussion and conclusions
This work reports the strain rate dependency of failure stress and
strain for composites made of basalt fibres and epoxy resin, produced as
laminates with quasi-isotropic [0∘, 45∘,þ45∘,90∘]S and orthogonal
[þ45∘, 45∘]4 layups. Three different strain rates in the range 10 3 –
3 � 102 are considered. Data at the highest strain rate is obtained using a
Split-Hopkinson Tension Bar. Our data is of sufficient quality to conduct
a meaningful strain rate sensitivity analysis of strength, i.e., our mea­
surements indicate significant changes in strength – relative to the ex­
periments’ standard deviation – as a function of strain rate. We quantify
the strain rate sensitivity of strength as approximately 3.5% per order of
magnitude of loading rate, both for quasi-isotropic and �45∘ specimens.
This is in agreement with existing high-quality studies on comparable
carbon and glass fibre epoxy composites in tension [8–13] and also
similar to what is observed in compression [14,20,21]. Our results
contradict the findings of a recent study on a similar basalt/epoxy
composite [7], who report a much more pronounced strain rate sensi­
tivity. In their study, a doubling of strength for a strain rate increase
from 10/s to 300/s was measured using a servo-hydraulic universal
testing machine. We argue that the Split-Hopkinson Tension Bar should
be used as the standard method for testing dynamic behaviour of com­
posites: Whilst servo-hydraulic machines are detrimentally affected by
pronounced wave propagation effects on short time scales, the Hop­
kinson Bar method incorporates this very phenomenon as the basis for
its measuring principle. To facilitate the use of the Split-Hopkinson
Tension Bar for composites, we introduce a novel system for easily
mounting specimens of strip shape to the Hopkinson bars. This new
mounting system provides constant acoustic impedance and does not
tivity is nearly identical for quasi-isotropic and �45∘ laminates. This is
somewhat surprising, as �45∘ specimens load the thermoplastic and thus
strain rate sensitive polymer matrix in shear, whereas much of the load
in the quasi-isotropic specimens is carried by the 0∘ plies, which are not
expected to exhibit pronounced strain rate sensitivity due to their
mineral constitution. To further investigate this issue, future work
should address accurate testing of 0∘ specimens with Split-Hopkinson
methods. This, however, is challenging because i) such specimens
require much stronger holding forces than the laminates considered
here, and ii) the small failure strain reduces the time available for
equilibration such that accurate measurements might prove difficult.
Funding
This work was supported by the Gips-Schüle-Stiftung and the Carl-
Zeiss-Stiftung [Sonderlinie 2017/2018 Grundlagenwissenschaften mit
Anwendungsbezug].
G.C. Ganzenmüller et al. Composites Part B 171 (2019) 310–319

Appendix A. Example Split-Hopkinson High Speed Images

Fig. A11. Sequence of high-speed images showing the deformation behaviour of a quasi-isotropic basalt-fibre/epoxy composite test in our Split-Hopkinson Tension
Bar at a loading rate of 300 /s.

References [6] May M, Hesebeck O, Marzi S, B€ ohme W, Lienhard J, Kilchert S, Brede M,

Hiermaier S. Rate dependent behavior of crash-optimized adhesives – experimental
characterization, model development, and simulation. Eng Fract Mech 2015;133:
[1] Kogan FM, Nikitina OV. Solubility of chrysotile asbestos and basalt fibers in
112–37. https://doi.org/10.1016/j.engfracmech.2014.11.006.
relation to their fibrogenic and carcinogenic action. Environ Health Perspect 1994;
[7] Chen W, Hao H, Jong M, Cui J, Shi Y, Chen L, Pham TM. Quasi-static and dynamic
102(Suppl 5):205–6.
tensile properties of basalt fibre reinforced polymer. Compos B Eng 2017;125:
[2] De�ak T, Czig�any T. Chemical composition and mechanical properties of basalt and
123–33. https://doi.org/10.1016/j.compositesb.2017.05.069.
glass fibers: a Comparison. Textil Res J 2009;79(7):645–51. https://doi.org/
[8] Gilat A, Goldberg RK, Roberts GD. Experimental study of strain-rate-dependent
10.1177/0040517508095597.
behavior of carbon/epoxy composite. Compos Sci Technol 2002;62(10):1469–76.
[3] Liu Q, Shaw MT, Parnas RS, McDonnell A-M. Investigation of basalt fiber composite
https://doi.org/10.1016/S0266-3538(02)00100-8.
mechanical properties for applications in transportation. Polym Compos 2006;27
[9] Eskandari H, Nemes JA. Dynamic testing of composite laminates with a tensile split
(1):41–8. https://doi.org/10.1002/pc.20162.
Hopkinson bar. J Compos Mater 2000;34(4):260–73. https://doi.org/10.1177/
[4] Jamshaid H, Mishra R. A green material from rock: basalt fiber – a review. J Text
002199830003400401.
Inst 2016;107(7):923–37. https://doi.org/10.1080/00405000.2015.1071940.
[10] Daliri A, Vijayan A, Ruan D, Wang CH. High strain rate tensile properties of basalt-
[5] Wu JQ, Zhong ZL. The applications of basalt fiber in the automotive industry in the
fibre reinforced polymer composites. In: 17th Australian international aerospace
future. Adv Mater Res 2011;(332–334):723–6. https://doi.org/10.4028/www.
congress : AIAC 2017; 2017. p. 164.
scientific.net/AMR.332-334.723.

318
G.C. Ganzenmüller et al.
[11] Spronk S, Gilabert Villegas FA, Sevenois R, Garoz G� omez D, Van Paepegem W.
Tensile rate-dependency of carbon/epoxy and Glass/polyamide-6 composites. In:
ECCM18 - 18th european conference on composite materials; 2018.
[12] G�omez-del Río T, Barbero E, Zaera R, Navarro C. Dynamic tensile behaviour at low
temperature of CFRP using a split Hopkinson pressure bar. Compos Sci Technol
2005;65(1):61–71. https://doi.org/10.1016/j.compscitech.2004.06.004.
[13] Jacob GC, Starbuck JM, Fellers JF, Simunovic S, Boeman RG. Strain rate effects on
the mechanical properties of polymer composite materials. J Appl Polym Sci 2004;
94(1):296–301. https://doi.org/10.1002/app.20901.
[14] Daniel IM, Werner BT, Fenner JS. Strain-rate-dependent failure criteria for
composites. Compos Sci Technol 2011;71(3):357–64. https://doi.org/10.1016/j.
compscitech.2010.11.028.
[15] Zhang H, Yao Y, Zhu D, Mobasher B, Huang L. Tensile mechanical properties of
basalt fiber reinforced polymer composite under varying strain rates and
temperatures. Polym Test 2016;51:29–39. https://doi.org/10.1016/j.
polymertesting.2016.02.006.
[16] Ramesh KT. High rates and impact experiments. In: Springer handbook of
experimental solid mechanics. Boston, MA: Springer; 2008. p. 929–60. https://doi.
org/10.1007/978-0-387-30877-7.
[17] Eriksen RNW. Of DCAMM special reports. High strain rate characterisation of
composite materials, vol. 179. DTU Mechanical Engineering; 2014.
[18] Chen WW, Song B. Split Hopkinson (Kolsky) bar: design, testing and applications.
Springer Science & Business Media; 2010.
[19] Wang W, Makarov G, Shenoi RA. An analytical model for assessing strain rate
sensitivity of unidirectional composite laminates. Compos Struct 2005;69(1):
45–54. https://doi.org/10.1016/j.compstruct.2004.04.017.
[20] Ochola R, Marcus K, Nurick G, Franz T. Mechanical behaviour of glass and carbon
fibre reinforced composites at varying strain rates. Compos Struct 2004;63(3–4):
455–67. https://doi.org/10.1016/S0263-8223(03)00194-6.
[21] Hsiao HM, Daniel IM, Cordes RD. Strain rate effects on the transverse compressive
and shear behavior of unidirectional composites. J Compos Mater 1999;33(17):
1620–42. https://doi.org/10.1177/002199839903301703.
319
Composites Part B 171 (2019) 310–319
[22] Foster J. Comments on the validity of test conditions for Kolsky bar testing of
elastic-brittle materials. Exp Mech 2012;52(9):1559–63. https://doi.org/10.1007/
s11340-012-9592-6.
[23] Harding J, Welsh LM. A tensile testing technique for fibre-reinforced composites at
impact rates of strain. J Mater Sci 1983;18(6):1810–26. https://doi.org/10.1007/
BF00542078.
[24] Ganzenmüller GC, Langhof T, Hiermaier S. A constant acoustic impedance mount
for sheet-type specimens in the tensile split-hopkinson bar. EPJ Web Conf 2018;
183. https://doi.org/10.1051/epjconf/201818302064. 02064.
[25] Plappert D. Quasi-static testing of basalt-fibre reinforced epoxy composites
according to DIN ISO 527, Tech. rep., to be published elsewhere. 2018.
[26] Ganzenmüller GC, Blaum E, Mohrmann D, Langhof T, Plappert D, Ledford N,
Paul H, Hiermaier S. A simplified design for a split-hopkinson tension bar with long
pulse duration. Procedia Engineering 2017;197:109–18. https://doi.org/10.1016/
j.proeng.2017.08.087.
[27] Guizar-Sicairos M, Thurman ST, Fienup JR. Efficient subpixel image registration
algorithms. Opt. Lett., OL 2008;33(2):156–8. https://doi.org/10.1364/
OL.33.000156.
[28] Frew DJ, Forrestal MJ, Chen W. A split Hopkinson pressure bar technique to
determine compressive stress-strain data for rock materials. Exp Mech 2001;41(1):
40–6. https://doi.org/10.1007/BF02323102.
[29] Kimberley J, Ramesh KT. The dynamic strength of an ordinary chondrite.
Meteoritics Planet Sci 2011;46(11):1653–69. https://doi.org/10.1111/j.1945-
5100.2011.01254.x.
[30] Yazici M. Loading rate sensitivity of high strength fibers and fiber/matrix
interfaces. J Reinf Plast Compos 2009;28(15):1869–80. https://doi.org/10.1177/
0731684408090379.
[31] Amos Gilat, Goldberg Robert K, Roberts Gary D. Strain rate sensitivity of epoxy
resin in tensile and shear loading. J Aerosp Eng 2007;20(2):75–89. https://doi.org/
10.1061/(ASCE)0893-1321(2007)20:2(75).
[32] Werner BT, Daniel IM. Characterization and modeling of polymeric matrix under
multi-axial static and dynamic loading. Compos Sci Technol 2014;102:113–9.
https://doi.org/10.1016/j.compscitech.2014.07.025.