MMS5 Case Study 3 – Marine Composites
PREDICTING THE PERFORMANCE AND FAILURE OF THICK COMPLEX CARBON FIBRE LAMINATES
BARRY SHEPHEARD, BUBA KEBADZE, DAN
JACKSON & MARK DIXON
DeepSea Engineering and Management
Ltd, Epsom, Surrey KT17 1BL
Predicting failure of thick
complex laminates using analytical and numerical tools in the past has been
fraught with error. Using conventional analytical formulations, as well as
being poor predictors of failure, they tell us nothing about the part
performance as load is applied and provides no indication of the mechanisms for
failure. For thick complex structures experiencing complex load patterns it is
necessary to develop numerical tools that provide designers with information on
how an as built part would behave during its normal operation, where the likely
failure initiators are and what mechanism leads to failure. DeepSea Engineering
has developed progressive failure models for thick complex laminates. As built
samples were tested to obtain their load displacement behaviour and were
compared to the model predictions based on coupon data. Materials utilisation
factors were then applied to the coupon data in the progressive failure model
to correlate the load displacement curves with the test data. The correlated
results were then used to predict the performance of the as-built full-scale
component.
1. INTRODUCTION
Consider two carbon fibre/resin
combinations that were used to make similar but not identical parts. In this
example wet filament winding process and pre-preg filament winding process were
used to manufacture similar but not identical cylindrical parts. It was
expected that the parts in-service performance could be determined from an
artefact of the part using a 4-point bend test. The strength of both materials
was known from coupon test data. A virtual 4-point bend model was constructed
using a finite element (FE) model and the results compared to the actual test.
However, when both parts were modelled using the material (coupon) strength
data with a 100% utilisation of the material properties then the strength of
the part undergoing a virtual four point bend was over-predicted. Obviously the model can be changed using
lower material utilisation factor, which is simply a multiplier of the material
strength, to match the strength of the part. The methodology described in this
paper compares different size and strengths of curved artefacts subjected to
4-point bending, the correlation of a failure model with the test data and the
application of utilisation factors from the correlated model to predict the 4
point bend test results.
2. FOUR POINT BEND TEST
This section describes the test
set-up. To clarify, the 4-point bend test was selected for three reasons;
1. It was a sensible physical size
for direct comparison with FE such that the FE model would not become onerously
time consuming to run. The FE model was built in ABAQUS (version 6.4). Each
layer, its thickness and orientation was represented.
2. It was expected that the
results from the 4-point bend test could be applied to other models in order to
predict the performance of the part.
3. The 4-point bend test was undertaken
on artefacts from the parts themselves so it was expected that manufacturing
quality would de-facto be included in the test data.
Note that void volume, fibre
volume fraction and Tg test were also carried out to establish
manufacturing quality and provided a valuable mechanism for quantifying the
effects of manufacturing quality of part performance.
Figure 1 Left, Example of the test
set-up. Right, FE model of the test set-up Figure 1 left, clearly shows the
test set-up used an the artefact by the National Physical Laboratory in the UK.
Figure 1 right, shows the test set-up replicated in FE.
3. COMPARISON OF THE TEST AND FE
PREDICTION
This section presents the results
of two different material types and geometries. Type A was a wet filament wound
process with thickness `a' and Type B was a pre-preg filament wound process of
thickness `b'. The laminate orientation and layer thicknesses were also
different for each type. The analysis of the results of the two models is
presented below.
Figure 2 show the images of the
failure obtained during the test of type A, failure propagations in beams from
the virtual FE tests and the load displacement curve of the test and the FEA
model respectively.
Figure 2 Top Left, Image of
failure in Type A, Top Right and Bottom Left, Predicted failure propagation in
type A. Bottom Right, Load displacement curve for Type A
Figure 3 show the images of the
failure of obtained during the test of type B, failure propagations in beams
from the virtual FE tests and the load displacement curve of the test and the
FEA model respectively.
Figure 3 Top, Image of failure in Type B. Left, Predicted failure
propagation in type B. Right, Load displacement curve for Type B.
The failure propagation of the
virtual test of type A is shown in Figure 2 top right & bottom left. It can
clearly be seen to correlate well with actual test shown in Figure 2 top left.
The delaminations at the mid point of the test piece and around 1/3rd of the
thickness are clearly identified, as is the relative severity of the
delaminations. Figure 2 bottom right illustrates the difference between the
predicted load displacement curve and that of the actual test using 100%
materials utilisation. While the displacement to failure correlates well, the
actual load seen in the virtual test is much higher. Failure of type B was
concentrated in the top layers as can be seen in the top image in Figure 3.
This is the scenario predicted by FEA as the virtual test also shows failure
dominant along the top layers, Figure 3 bottom left. The load displacement
curves of the tests and the FE models for the beams for type B using 100%
stiffness and strength utilisation factors, as shown in Figure 3 bottom right.
It can be seen that even though the response of the beams were captured
qualitatively the magnitudes of the stiffness and the strength are over
predicted.
4. CORRECTION FACTORS
In order to make comparison
between the two samples the models were tuned to match the stiffness and
strength obtained from the experiment. This was achieved by using lamina
property utilisation factors on stiffness and strength, kE and kF
respectively. The material properties used in the model were then obtained by
multiplying elastic and strength properties of lamina by appropriate utilisation
factors. Figure 4 show the load-displacement curves for the artefacts where the
finite element model was adjusted to incorporate the material property
utilisation factors. It can be seen that the use of simple coefficient on the
material properties brings the prediction in to line with the actual test.
Figure 4 Left, displacement curves
for the type A. FEA model uses degraded properties. Right, Load displacement
curves for the type B. FEA model uses degraded properties.
5. SELECTION OF THE FAILURE CRITERIA
An important consideration was the
selection of an appropriate failure criterion. For the above example Hashin's
criteria was used (1) was used. This was selected as it is perceived to be the most
appropriate for the circumstances associated with the test since Hashin has the
advantage of being able to differentiate fibre and matrix failure and identify
compressive or tensile modes. A thorough review of the failure criteria was
then undertaken, there are some useful review papers in this field in
particular those of Soden et al (2), Kaddour et al (3), Puck & Schurmann
(4), Hinton & Soden (5) and Paris (1). Hashin and Puck’s theories
differentiate the failure mechanism associated with the fibres and matrix and
also distinguishes between tension and compression. The primary difference
between these theories and the other theories such as Tsai is that Tsai for
example will predict laminate failure but not distinguish between fibre or
matrix compression or tension. As seen in the reference papers there are
numerous failure criteria and it was not the purpose of this study to provide a
comprehensive evaluation, merely to review them and select notable examples for
comparison with the 4 point bend test, its failure modes and mechanism. To that
end Hashin was an appropriate selection because of its potential to provide
information on the mode of failure and Tsai was chosen as a comparison because
it is a well establish and extensively used criteria. Four identical tests were
conducted and both Hashin and Tsai compared.
Figure 5 Left, Comparison of
Hashin and Tsai with four identical bend test using identical utilisation
factors. Top Right, Hashin failure propagation. Bottom Right, Tsai Failure
propagation
The comparison showed that both
Hashin and Tsai could be used to as appropriate techniques. As stated earlier
the benefit of Hashin was the ability to differentiate the modes, this can be
more clearly illustrated in Figure 5. Using type A as benchmark Figure 2 top
left showed failure at the mid point through the thickness. Tsai does not
predict this, see Figure 5 bottom right, whereas Hashin does, see Figure 5 top
left. In addition Hashin also predict the damage done by the loading points and
the failure roughly 1/3rd through the thickness.
6. CONCLUSIONS
DeepSea Engineering has developed
progressive failure models for complex laminates and a process of validation
testing as-built artefacts of the part and compared with the model predictions.
The model predictions were then modified to follow the as-recorded data.
Materials utilisation factors were then applied to give a measure of the
performance of the part compared to coupon data.
It was seen that the failure
criteria best suited to matching the actual failure modes and mechanism was the
Hashin criteria.
7. ACKNOWLEDGEMENTS
The authors a pleased to
acknowledge the testing undertaken at the following facilities; National
Physical Laboratory (NPL) in the UK, Composites Testing Laboratory (CTL) in
Ireland and at EADS Composites Atlantic in Canada. The authors also acknowledge
Cooper Cameron Corporation in the US for authorising its publication
REFERENCES
1. F. Paris, A Study of
Failure Criteria of Fibrous Composite Materials, NASA/CR2001-210661, March 2001
2. P.D. Soden, M.J. Hinton
and A.S. Kaddour, A Comparison of the Predictive Capabilities of Current
Failure Theories for Composite Laminates, Composites Science and Technology,
58, 1225-1254, 1998
3. A.S. Kaddour, M.J.
Hinton and P.D. Soden, A Comparison of the Predictive Capabilities of Current
Failure Theories for Composite Laminates: Additional Contributions, Composites
Science and Technology, 64, 449-476, 2004
4. A. Puck & H.
Schurmann, Failure Analysis of FRP Laminates by Means of Physically Based
Phenomenological Models, Composites Science and Technology 62, 1633-1662, 2002.
5. M.J. Hinton & P.D.
Soden, Predicting Failure in Composite laminates: The Background to the
Exercise, Composites Science and Technology 58, 1001-1010,1998