Author's personal copy

Composites: Part A 41 (2010) 888–901

Contents lists available at ScienceDirect

Composites: Part A
journal homepage: www.elsevier.com/locate/compositesa

Defect and damage analysis of advanced discontinuous carbon/epoxy

composite materials
Paolo Feraboli a,*, Tyler Cleveland b, Marco Ciccu b, Patrick Stickler c, Luciano DeOto d
a
Automobili Lamborghini Advanced Composite Structures Laboratory, Department of Aeronautics and Astronautics, Box 352400, Guggenheim Hall,
University of Washington, Seattle, WA 98195-2400, United States
b
Department of Aeronautics and Astronautics, University of Washington, Seattle, WA, United States
c
Boeing 787 Technology Integration, Boeing Commercial Airplanes, The Boeing Co., Everett, WA, United States
d
Advanced Composites Development Center, Automobili Lamborghini S.p.A., Sant’Agata Bolognese, Italy 40019, Italy

a r t i c l e i n f o a b s t r a c t

Article history: Recent composite technology research and development efforts have focused on discontinuous carbon
Received 8 August 2009 fiber/epoxy molding systems derived from chopped aerospace-grade unidirectional tape prepreg. This
Received in revised form 2 March 2010 study analyzes in detail the meso-structure of this class of materials, which exhibit point-to-point vari-
Accepted 3 March 2010
ations associated with the random chip distribution, by means of destructive and non-destructive inspec-
tions, in the attempt to identify characteristic traits that can yield insight in its quality and performance.
Results show that several types of defects can be encountered within the molded panel, such as macro-
Keywords:
voids, fiber kinking and swirling, or resin-rich areas. However, it is found that failure may or may not
A. Discontinuous reinforcement
E. Compression moulding
occur in proximity of these hot spots, independently from their size and location, even for specimens con-
A. Carbon fiber taining the circular hole. Therefore it appears that for this class of materials conventional ultrasonic
B. Defects inspection and defect classification may not be suitable as criteria for part acceptance or rejection.
Ó 2010 Elsevier Ltd. All rights reserved.

1. Introduction tems is higher than the UD prepreg from which they are derived,
their suitability to be molded in complex geometries with lower
Recent composite technology research and development efforts manufacturing costs and at higher rates can justify their adoption
have focused on new out-of-autoclave material forms, and auto- to reduce overall part acquisition costs.
mated processes that can markedly increase production efficien- The authors have been performing fundamental research on
cies. The interest of the aerospace community for short fiber these materials, and previous experimental results have been re-
composites, such as Sheet Molding Compounds (SMC), dates back ported [6–8]. This class of materials, which sits between traditional
to the 1960s and the pioneering work of Halpin, Pagano and Kardos SMC and prepreg tape, has shown excellent mechanical properties
[1–3]. In recent years, airframe manufacturers have been propos- [5,6], particularly for stiffness-dominated designs, since the aver-
ing the use of high-performance discontinuous systems that are age modulus reported can be as high as that of the reference qua-
suitable for compression molding of primary structures. Commer- si-isotropic continuous tape laminate. Furthermore, while tensile
cial applications for this type of material form already exist, strength is less than half that of the reference quasi-isotropic con-
although using different resin systems and fiber types and lengths, tinuous tape laminate, compressive and shear strengths are much
under various manufacturers and trade names (e.g. Quantum Lytex closer to those of the reference tape laminate. If material and pro-
4149 and Hexcel HexMCÒ). The Boeing 787 Dreamliner for exam- cess are accurately controlled, the variation observed in the mea-
ple makes use of AS4/8552 HexMCÒ for the window frames [4,5], sured strength is also relatively low around 8%, and competitive
as well as other primary and secondary structural elements. These with the 4–6% observed for the tape precursor. Results also show
mats are processed by consolidating sheets of randomly-distrib- that the distribution of chips is indeed random, yielding in-plane
uted ‘‘chips”, up to 2.0 in. long and 0.33 in. wide (50.8 mm  quasi-isotropic elastic and strength properties. These statements
8.4 mm) in the form of a roll. These chips are obtained from have been verified by testing coupons obtained at 0°, 45° and 90°
pristine aerospace-grade unidirectional (UD) carbon fiber/epoxy directions.
prepreg, which is first slit longitudinally and then chopped. These materials also pose unique challenges for engineers
Although the raw material cost associated with these chopped sys- tasked with design, analysis and certification of primary structural
parts manufactured with them. In particular, two unique charac-
* Corresponding author.
teristics have been reported in [7,8] respectively. The open-hole
E-mail address: [email protected] (P. Feraboli). tension strength of these materials has been shown [7] to be

1359-835X/$ - see front matter Ó 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compositesa.2010.03.002
 Author's personal copy

P. Feraboli et al. / Composites: Part A 41 (2010) 888–901 889

virtually notch-insensitive at the macroscopic level, possibly due

to the presence of internal stress concentrations arising from the
heterogeneous nature of the sub-structure. This sub-structure
can be identified as a meso-structure rather than a micro-struc-
ture, since it appears to be associated to the size and properties
of the chips, rather than to the characteristics of the constitutive fi-
ber and resin phases. An ‘‘inherent material” stress concentration
factor is derived, and used to explain the tendency of the material
to often fail at the gross-section regardless of the presence of a
hole. This behavior is hole size-dependent and it transitions to
exclusive net-section failure for large hole sizes. Common analysis
methods used for homogenous materials, independent from their
orthotropic nature, cannot be used to predict notched strength
load and location of failure, although it was shown that a modified
Point Stress Criterion employing large characteristic length values
could be used. The non-homogeneous meso-structure appears to
be responsible also for the observations in [8], where axial Young’s
modulus measurements, conventionally obtained by strain gage or
extensometer, yield up to 19% Coefficient of Variation (CoV) from
specimen to specimen as well as within a single specimen. A series
of tensile tests was conducted while systematically varying strain
gage length, from 0.125 in. up to 2.0 in. (3.2–50.8 mm), as well as
strain gage and extensometer location along the length and across
the width of the specimen. Values have been measured in differ-
ent locations on the same specimen that vary from as low as Fig. 1. Molded and trimmed panel.
3.0 Msi to as high as 9.0 Msi (20.68–62.04 GPa). Longer strain
gages do not appear to yield better results than shorter gages,
and extensometer readings are not necessarily better than strain A total of 24 specimens are evaluated, as summarized in Table 1,
gage ones, thus suggesting that the length scale associated with and include a broad range of physical and mechanical property
these strain/modulus variations is greater than what can be mea- evaluations. Non-destructive evaluation (NDE) is performed by
sured with a large strain gage. The use of Digital Image Correla- means of pulse-echo C-scan ultrasound, a Physical Acoustics Corp.
tion technique (DIC) to obtain full-field strain distributions on UltraPAC NDT Automation immersion system, using a 5 MHz sen-
the surface of the specimen revealed a complex state of variable sor. The multichromatic scale ranges from red, which indicates no
strain in the specimen, with peaks and valleys throughout the signal loss, to blue,1 or complete loss of signal, Fig. 3. The signal
specimen. It also allowed for obtaining the most comprehensive threshold is set at 6 dB loss, which corresponds to 80% of the emit-
modulus measurement by averaging the strain values over the ted signal. As it can be seen, the bulk of the laminate is comprised
entire surface of the tensile coupon, and thus showing that the of a diffused mix or red, orange and yellow areas, which indicate a
average material modulus is indeed relatively constant with 5% highly heterogeneous structure. C-scan images are difficult to
CoV. These observations support that the observed variation in lo- interpret compared to tape laminates. With ultrasonic inspection,
cal modulus are peculiar to this material, whose heterogeneous any change in material properties, such as the presence of a defect
meso-structure is comprised of highly orthotropic and long or of a thickness transition, is characterized by attenuation in the
reinforcements. signal. For this material form, there are point-to-point variations
in material properties due to the random chip distribution, which
generates signal attenuation often not corresponding to the pres-
2. Material fabrication and test setup ence of a true defect. It becomes difficult to isolate the presence
of a small defect from the background noise. However, a few small
All discontinuous carbon/epoxy specimens are manufactured in regions are indicated with green and blue, and are indicative of re-
the laboratory starting from unidirectional tape prepreg. The system gions where the signal attenuation is greater than the surrounding
is a 350 °F cure (177 °C), designated for vacuum bag, autoclave cure, material. These regions are here defined as ‘‘hot spots”, or regions
and has a resin content of 45%. During molding, 2–3% by volume or of interest, as it is not possible to define them as true defects with-
resin is lost in flash and through vent holes, which are used to evac- out further characterization.
uate air, so that the final fiber volume content is around 57–78%. Pulsed thermography is used to evaluate its suitability to better
The chip dimensions are 2.0 in. long  0.33 in. wide (50.8 mm  detect the presence of damage. Thermal Wave Imaging, Inc. is sub-
8.4 mm). Detailed discussion of the compression molding manufac- contracted to perform non-destructive inspection of a selected
turing process for the material is reported in [6]. After extraction number of panels using their Echotherm Infrared Flash Thermogra-
from the mold, the panel is trimmed 1.0 in. (25.4 mm) on all sides phy system. The image of Fig. 4 shows that for this material system
to remove the regions of high flow toward the mold edges, where thermography is not a suitable tool to detect the presence of small
it is more likely that chip alignment and distortion may occur, manufacturing defects, or hot-spots. Using the 5 MHz frequency as
Fig. 1. Although this is a low-flow process, unlike typical SMC mold- the benchmark for ultrasonic inspection, sensor frequencies of 2.25
ing where the charge covers only a small fraction of the mold, some and 10 MHz are evaluated to assess their ability to isolate the pres-
degree of resin flow still takes place during the molding process. This ence of hot spots from the-background noise, Fig. 5. It can be seen
is not sufficient to create a noticeable alignment of the fibers along that at 10 MHz the image shows a greater number of regions of sig-
any preferred direction, but inevitably lead to the formation of small nal attenuation, but the increased sensitivity and detail is hindered
resin-rich areas. Average panel thickness is 0.165 in. (4.2 mm), and
corresponds to a nominal number of 32 chips through the thickness, 1
For interpretation of color in Figs. 3, 8 and 10, the reader is referred to the web
each with a cured thickness of 0.005 in. (0.13 mm), Fig. 2. version of this article.
 Author's personal copy

890 P. Feraboli et al. / Composites: Part A 41 (2010) 888–901

Fig. 2. Typical cross section of pristine (high quality) material, showing the randomly oriented chips alternating through the thickness.

Table 1
Summary of specimens evaluated in this study.

Specimen # NDE Final operation Ultimate strength [ksi] (MPa)

1 C-scan/thermography Two hot-spot regions microscopy –
2 C-scan Three hot-spot regions microscopy –
3 C-scan/thermography Two hot-spot regions microscopy –
4 C-scan Four hot-spot regions microscopy –
5 C-scan Three regions of acid digestion –
6 C-scan Three regions of acid digestion –
7 C-scan Surface machining –
8 C-scan Surface machining –
9 C-scan before and after failure Load to ultimate failure 45.4 (313)
10 C-scan before and after failure Load to ultimate failure 41.3 (285)
11 C-scan before and after failure Load to ultimate failure 46.4 (320)
12 C-scan before and after failure Load to ultimate failure 49.5 (341)
13 Progressive C-scan up to failure Progressive loading up to failure 44.4 (306)
14 Progressive C-scan up to failure Progressive loading up to failure 54.1 (373)
15 Progressive C-scan up to failure Progressive loading up to failure 41.2 (284)
16 C-scan before and after loading to 70% of ultimate Hot-spot growth microscopy –
17 C-scan before and after loading to 80% of ultimate Hot-spot growth microscopy –
18 C-scan before and after failure Load to ultimate failure in presence of open hole 46.6 (321)
19 C-scan before and after failure Load to ultimate failure in presence of open hole 41.1 (283)
20 C-scan before and after failure Load to ultimate failure in presence of open hole 44.3 (305)
21 C-scan before and after failure Load to ultimate failure in presence of open hole 43.6 (301)
22 Progressive C-scan up to failure Progressive loading up to failure in presence of open hole 39.5 (272)
23 Progressive C-scan up to failure Progressive loading up to failure in presence of open hole 45.9 (316)
24 Progressive C-scan up to failure Progressive loading up to failure in presence of open hole 36.3 (250)

by a greater difficulty to discriminate these regions from the back- ester and ending up with 1 h polishing with a non-nap polyester
ground noise. On the other hand, the 2.25 MHz sensor shows a cloth and 10% alumina solution. Magnification from 50 to 500
much smoother signal contour, and the areas of signal attenuation is used to detect the nature of the hot spots, and to observe the
are clearly defined. However, the image loses sufficient detail to presence of damage.
change the shape and size of some key regions of signal attenua- A selected number of specimens are machined from the molded
tion, and in some cases completely misses them. For these reasons, panel to smaller size, approximately 1.5 in.  3.0 in. (38.10 mm 
ultrasonic inspection using the 5 MHz sensor is deemed the most 76.20 mm). Following ultrasonic inspection to detect the presence
suitable NDE technique, giving the best compromise between sen- of hot spots, these are mounted on a milling machine and sub-
sitivity and noise, and will be used for the rest of this study. jected to further machining. From the top surface, the surface is
Micrographic coupons are extracted from molded panels in the milled at 0.0025 in. (0.06 mm) increments through the thickness,
regions where the ultrasound indicates the presence of a hot spot. corresponding to half the chip thickness. Since the ultrasonic im-
Since the panels have random chip distribution and have been age shows only a projected hot spot area, it is not easy to deter-
shown to be quasi-isotropic in the plane, the orientation of the mine its depth within the laminate. By removing ‘‘planes” of
cross section cut is not relevant. The specimen is first vacuum- material at each step it is possible to expose the underlying mi-
impregnated with mounting epoxy resin, then pressure-cured to cro-structure and observe by optical microscopy the nature and
minimize formation of air bubbles, and is finally polished for location of the hot spots.
inspection. The specimen is polished with a 6-step process: Six small specimens, having a weight of 1 g, are evaluated for fi-
180 grit, 600 grit, 1200 grit, 9-lm, 3-lm silk, 3-lm non-nap poly- ber volume content by means of ASTM standard D3171 for acid
 Author's personal copy

P. Feraboli et al. / Composites: Part A 41 (2010) 888–901 891

hot plate to boiling temperature. At that point 35 ml of hydrogen

peroxide are added to clarify the mixture, which indicates that
indicates that the resin is fully dissolved. While the epoxy matrix
dissolves in acid, the fibers are unaffected by the corrosive mixture.
The residue is cooled down in cold water and filtered in 15-lm
ceramic filters using a vacuum pump to separate the liquid, which
is the collected in a catcher. After desiccation, the left over fibers
are weighed to measure the remaining volume. The difference be-
tween the composite specimen volume before acid digestion and
the volume of the left over fibers gives a measure of the resin vol-
ume content. It should be emphasized that this process is very
slow and lengthy, and selection of the appropriate filters as a com-
promise between process time and measurement accuracy is key.
The burn-off test, which can be used to separate resin from fibers
by ignition loss in a furnace, is a much speedier process and less
operator intensive. Although preferable from a user perspective,
such test can be used only for glass fibers, since carbon fibers car-
bonize and vaporize at high temperatures.
From the molded panels, tensile specimens are machined to the
Fig. 3. C-scan pulse-echo ultrasonic image of a molded panel using a 5 MHz sensor.
Boeing standard test method for unnotched tension [10]. The spec-
imens have straight-sided rectangular geometry with dimensions
12.0 in.  1.5 in. (304.80 mm  38.10 mm). Since glass/epoxy fab-
ric tabs of length 2.0 in. (25.4 mm) are bonded to the specimen
using 3 M Scotchweld film adhesive, the effective gage length is re-
duced to 8.0 in. (203.2 mm). All specimens are loaded at a rate of
0.05 in./min (1.3 mm/min) in a 2-grip Instron hydraulic tension/
compression test frame. From previous research [6–8], the authors
had identified the average tensile failure strength of this material.
In this study, the specimens are loaded progressively to failure at
increments of 60%, 70%, 80%, 90% and 100% of the average ultimate
failure strength. After each loading step, the specimens are re-
moved and inspected via ultrasound to monitor the initiation of
damage and the growth of hot spots. After inspection, the speci-
mens are re-positioned in the test frame and the loading is in-
creased to the next step. For a selected number of specimens, the
load is interrupted at one of the step and the specimen is sacrificed
for destructive inspection. Microscopy is used to evaluate the nat-
ure of the damage and its extent.

Fig. 4. Pulsed thermography analysis of molded panel in Fig. 3, courtesy of Thermal

Wave Imaging Inc. 3. Defect inspection and characterization

Ultrasonic C-scan inspection is used on specimens #1–4 to de-

tect the presence and location of hot spots, but only two specimens
are shown in this section, as they are sufficiently representative of
the whole set. At the location of the hot spots, the specimens are
sectioned to obtain smaller specimens for micrographic inspection.
Specimen #2 is shown in Fig. 6, and highlights the presence of
three well-defined regions of signal attenuation, indicated in blue.
The micrographic images associated with each of the hot spots are
shown on the right. At position 1, the hot spot is associated with
the presence of a large swirl, which is a region of extensive chip
kinking and distortion associated to a region of high flow. As men-
tioned previously, although these materials are designed to be low-
flow, that is the majority of the mold cavity is covered by the
charge, in some regions higher flow may occur due to an inconsis-
tency in the charge distribution. Positions 2 and 3, in the same
specimen, show large resin pockets, one toward the surface and
the other toward the midplane of the laminate. During molding
Fig. 5. C-scan pulse-echo ultrasonic image of a specimen using 2.25, 5 and 10 MHz the resin liquefies and fills the interstices left between neighboring
sensors (top to bottom). chips. While this is a desired attribute, if the mat used in the charge
exhibits a particularly large empty space, for example associated
with a missing chip, the resin will concentrate in this region and
digestion [9]. After measuring the volume of the composite speci- form resin-rich pockets. Specimen #4, Fig. 7, exhibits two hot
men by immersion in distilled water, the specimens are immersed spots, which both reveal to be associated to the presence of several
in a beaker containing a 20 ml of sulphuric acid and heated on a large voids. These macro-voids are much larger than the typical
 Author's personal copy

892 P. Feraboli et al. / Composites: Part A 41 (2010) 888–901

Fig. 6. Micrographic images of three different hot-spot regions in the specimen #2 shown on the left.

Fig. 7. Micrographic images of two different hot-spot regions in the specimen #4 shown on the left.

voids associated with the diffused porosity encountered in auto- Specimens #7 and #8 are first inspected by C-scan to reveal the
clave processing, and form during manufacturing when air bubbles presence of hot spots, whose nature location through the thickness
are entrapped in the laminate. Because of the high pressures in- is not discernible based on the ultrasonic image alone. They are
volved in manufacturing these materials, ranging from 750 to then placed in a milling machine and subjected to material re-
1200 psi (51.7 and 82.8 bars respectively), porosity is not a typical moval from the top surface down. This process can be performed
problem. Typical prepreg autoclave cure is performed at much across the entire surface of the specimen at the same time, or at
lower pressures, ranging from 80 to 100 psi, or 5.5 and 6.9 bars subsequent steps, Fig. 9A. By superimposing the C-scan image to
respectively. On the other hand, because the pressures are so high the machined surface it is possible to focus on the locations of
and the curing time so low (approximately 5–20 min in the press, the hot spots, Fig. 9B. After each pass, which is associated with
compared to the several hours used for autoclave curing), any air the removal of half a chip thickness, the specimen is inspected un-
bubble entrapped in the charge will not have a chance to escape. der the microscope to verify if the hot spot is visible. From the sur-
It should be noted that one of the two hot spots of specimen #4, face of the specimen in Fig. 9C it is possible to see the presence of
the one on the lower left in Fig. 7, corresponds to a small mark macro-voids, visible as either small concave regions or striations,
on the outer surface of the panel. Such defects, that are detectable which indicate an elongated shape. The material removal opera-
by visual inspection, have the advantage that can be accounted for tion is very time-consuming, and does not yield particular insight
during quality acceptance screening. on the nature of the defects, although it does supply an interesting
Specimens #5 and #6 are used for acid digestion testing. For view of the meso-structure of the material along planes parallel to
each panel, three specimens are identified by C-scan as corre- outer surfaces. Since the chips do not remain perfectly aligned dur-
sponding to a region of pristine signal (red/orange), of partial signal ing the molding process, this structure is characterized by highly
attenuation (green/light blue), and of high signal attenuation (dark heterogeneous collection of pseudo-elliptical regions having differ-
blue), Fig. 8. Results show that the pristine regions have high fiber ent chip orientations, Fig. 9D.
volume content of 59%, regions of modest attenuation have an
average of 55.5%, while regions of high attenuation have an aver- 4. Damage growth and failure
age of 53.7%. These results confirm the microscopic analysis obser-
vations, which seem to indicate that these hot spots are associated Specimens #9–12 are selected based on ultrasonic inspection
to macro-voids, chip swirls or resin-rich areas, all of which are that contain several hot-spot regions, with the intent to see how
characterized by low fiber volume content. these may affect failure load and location. The hot spots vary in
 Author's personal copy

P. Feraboli et al. / Composites: Part A 41 (2010) 888–901 893

Fig. 8. Acid digestion locations and fiber volume content results for specimens #5 and #6.

size, location and intensity, as seen in Fig. 10, and include some behavior, and some of the larger, more intense ones may not grow
that are of high intensity (dark blue) but relatively isolated, and at all.
others that are less intense (light blue) but more diffused through- In order to better capture the evolution of damage and the
out the specimens. The larger hot spots are then traced out on the growth of these hot spots, three tensile specimens are loaded at
surface of the specimens by means of a white marker to monitor if increments up to catastrophic failure. After each loading incre-
failure occurs at these locations, Fig. 11. The specimens are then ment, the specimen is removed from the test frame and inspected
loaded to ultimate failure and inspected by ultrasound again, via C-scan. By means of image analysis, the total ‘‘damage area” is
Fig. 10. Three of the four specimens failed in the gage section, while measured, which is defined as the area of the specimen exceeding a
one (specimen 15) failed in proximity of the tabs. Although speci- predefined threshold for signal attenuation. Progressive C-scan
men #9 contains two large hot spots toward the upper and lower images are shown for specimens #13–15 in Figs. 12–14. Each fig-
parts of the image, failure occurs at a milder hot spot, located to- ure shows six C-scan images, corresponding to six fractions of ulti-
ward the center gage of the specimen. Specimen #10 contains 8 mate load: 0% (pristine specimen), 60%, 70%, 80%, 90%, and 100%
mild hot-spot regions dispersed along the specimen, of which (ultimate load and failure). Specimen #13, Fig. 12, shows the pres-
one appears to be slightly more intense than others. Failure occurs ence of a large, intense hot spot prior to loading. Such hot spot does
in the center section in an area that is characterized by relatively not grow significantly during loading, and ultimately does not pre-
mild and spaced hot spots, while the larger hot spot seems unaf- cipitate failure in its proximity. Failure occurs on the right side of
fected in size and intensity. On the other hand, two of the milder the specimen, in an apparently pristine region, where few mild
hot spots, one toward the upper part of the specimen, and one to- hot-spot regions grow during loading and coalesce. Specimen
ward the lower part, have grown significantly in size and intensity #14, Fig. 13, shows the presence of several hot spot region of high
during loading. Specimen #11 fails at the tabs, in proximity of the signal attenuation, with the largest one toward the right of the
grips, but a closer look reveals that a relatively dense number of specimen and in proximity of other two slightly smaller regions.
hot spots exist in that region prior to loading. From the post-failure During loading all hot spots grow in size and intensity, and finally
ultrasound, it appears that these hot spots grow and coalesce dur- failure precipitates when the two hot spots on the right coalesce.
ing loading, leading to failure in that region close to the tabs. An- For specimen #15, Fig. 14, only four C-scan images are shown since
other large hot spot region has grown during loading toward the it fails prematurely at 90% of the average ultimate strength of the
bottom of the specimen, in a region that was apparently free of material. The specimen exhibits a small but intense hot spot in
hot spots, and at failure its size and intensity is nearly as large as its right side, and two areas of diffused but low-intensity signal
that where failure occurred. Specimen #12 contains 7 hot-spot re- attenuation, in the center and to the left of the specimen. Upon
gions, the majority of which are concentrated on the left side of the loading, the left-most region grows at a much faster rate than
specimen in Figs. 10 and 11. The largest one, in the shape of a T, is the other hot-spot regions, including the high intensity one on
located toward the center section of the specimen, and is in close the right, and eventually leads to failure. For all specimens the
proximity to another large hot spot. Failure occurs in that region growth is sudden, and at 60% and 70% of ultimate load the change
after those two hot spots grow and coalesce. It should be noted in damage area is virtually undetectable. For the three specimens
that toward the bottom of the specimen, two other hot spots, lo- tested, Fig. 15 shows that at 60% of ultimate load there is virtually
cated on the opposite sides of the centerline, have grown signifi- no change in the total area of signal attenuation, while at 70% the
cantly and started to coalesce. area is between 10% and 20% larger than the pristine value. Only at
Preliminary observations coming from these tests are that ulti- 80% of ultimate load the growth becomes evident, at approxi-
mate failure may manifest in proximity of the dominant hot spot, mately 50% more than the pristine value, and increases more rap-
but in at least 50% of the cases it may occur either in a region of idly at higher load values, reaching 150% at 90% of loading.
neighboring hot spots of lower intensity or even in a region of Specimens #16 and #17 are inspected by ultrasound prior to
apparently pristine ultrasonic area. Unlike continuous fiber com- being loaded to 70% and 80% of ultimate load respectively. They
posites, the intensity and size of the defect, or hot spot as it is here are then inspected by ultrasound again to verify the growth of
called, may not dictate the location of failure. Secondly, during the hot spots and initiation of damage, and are then sectioned
loading some of these hot spots grow in size, and neighboring ones for microscopic analysis to understand the damage mechanisms.
tend to coalesce. However, not all hot spots show the same growth Fig. 16 shows this process for specimen #16, from which three dif-
 Author's personal copy

894 P. Feraboli et al. / Composites: Part A 41 (2010) 888–901

Fig. 9. (A–D) Top view of specimen #7 following surface machining to expose underlying structure and hot spots.

ferent regions are sectioned. These regions, which are designated gion of intense signal attenuation (M2), and an intermediate atten-
M1–M3 and showed top to bottom in Fig. 16, include a region of uation region (M1). Section M2 shows a range of multiple defects
apparent pristine conditions and no signal attenuation (M3), a re- that appear to concentrate in the same region. These include sev-
 Author's personal copy

P. Feraboli et al. / Composites: Part A 41 (2010) 888–901 895

#9 #10 #11 #12

Before Before Before Before
After After After After

Fig. 10. Ultrasonic images of specimens #9–12 before and after failure.

aminations are the result of the applied load and are responsible
for the increase in total ‘‘damage” area, or area of signal attenua-
tion as measured by C-scan. Section M1, Fig. 19, of mild signal
attenuation exhibits the presence of several intra-chip cracks di-
rected perpendicular to the cross section. These cracks are typical
of the early stages of damage initiation and growth, and form in
a region without obvious manufacturing flaws or apparent hot
spots. Section M3, Fig. 20, which belongs to a bright red region of
no signal attenuation, shows a pristine micro-structure as the
one reported in Fig. 2. Similar observations from specimen #17
suggest that the relationship between large manufacturing defects,
such as voids and swirls, and damage initiation and propagation
appears to be all but obvious. Damage does not emanate and grow
from the pre-existing defects, although it forms in their proximity.
It is possible that the presence of defects may be responsible for lo-
cally lower mechanical properties, such as Young’s modulus whose
point-to-point variation is documented in [8], and in turn may cre-
ate high strain gradients in their proximity. These gradients could
then affect the local load redistribution and precipitate failure in
the pristine material in proximity of these defects. In any case,
the presence of hot spots, which are confirmed to be manufactur-
ing defects of varying nature, does not consistently precipitate fail-
ure in their proximity.
It was previously reported by the authors in [7] that this mate-
Fig. 11. Specimens #9–12 after failure. rial form is particularly notch-insensitive, and exhibits an open-
hole tensile strength that is unchanged with respect to the unnot-
ched strength. Furthermore, for 0.25-in. (6.35 mm) circular hole, it
eral macro-voids of varying sizes concentrated on the left side of was reported that at least 50% of the specimens fail away from the
the cross section, and a large swirl, or region of high flow and fiber open hole, in the gross section. It is thought that such behavior is
kinking in the center of the cross section, Fig. 17. These are present due to the presence of defects, or hot spots, that create material
before loading and do not show signs of growth. On the right side stress concentrations more dominant than the geometric one. To
of the specimen, Fig. 18, several inter- and intra-chip cracks can be validate this theory, a series of notched tensile tests are performed
observed through the thickness, as well as multiple large adjacent on specimens #18–21. Four specimens are selected carefully to en-
delaminations, which appear to generate in resin-rich regions in sure that no significant hot spots are present, as shown in the C-
the proximity of chip terminations. Both transverse cracks and del- scan images of Fig. 20, which are characterized by a uniform red
 Author's personal copy

896 P. Feraboli et al. / Composites: Part A 41 (2010) 888–901

Load increase

Fig. 12. Specimen #13 undergoing progressive loading and ultrasonic imaging up to ultimate failure.

Load increase

Fig. 13. Specimen #14 undergoing progressive loading and ultrasonic imaging up to ultimate failure.

(high signal) projected area. The specimens are then loaded to fail- Another set of notched specimens is tested but, unlike the pre-
ure, but three of the four specimens fail away from the hole, in the vious ones, these are selected to one or more hot spots, to under-
gross section area. From the ultrasonic images it is not clear why stand if the presence of a clearly defined hot spot can precipitate
failure precipitates at a specific point along the specimen, since failure in its proximity. Specimens #22–24 are loaded to failure
there are no obvious regions of signal attenuation. at multiple increments, and between each loading step they are re-
 Author's personal copy

P. Feraboli et al. / Composites: Part A 41 (2010) 888–901 897

Load increase

Fig. 14. Specimen #15 undergoing progressive loading and ultrasonic imaging up to ultimate failure.

moved and inspected via ultrasound to monitor the growth of

damage. As in previous tests for unnotched specimens, five C-scan
images are shown for each specimen, corresponding to 0% (un-
loaded), 50%, 70% 90% and 100% (ultimate failure) increments.
Specimen #22, Fig. 21, contains a relatively large hot spot in the
proximity of the open-hole. Upon loading, the hot spot grows
across the specimen width in a direction away from the hole,
and eventually leads to failure in the gross section, although close
to the hole. During loading, two large hot spots form on the left
side of the specimen, and at failure they have grown to the point
that they almost coalesce. Specimen #23, Fig. 22, contains a mildly
homogeneous signal, with no clearly defined hot spot. Upon load-
ing, four/five mild hot spots begin to form at different locations
along the specimen, and eventually two of them on opposite sides
of the specimen width coalesce and precipitate failure. Specimen
#24, Fig. 23, exhibits a high-intensity hot spot and a wide region
of diffused mild-intensity hot spots to the left of the specimen.
During loading these hot spots grow, and in particular the smaller
Fig. 15. Normalized damage area vs. normalized applied stress for the three
but more intense one coalesces with a neighboring one and leads
specimens of Figs. 12–14. to failure.

Fig. 16. Specimen #16, inspected by ultrasound before and after loading to 70% of ultimate strength, is then sectioned in three different locations M1–M3.
 Author's personal copy

898 P. Feraboli et al. / Composites: Part A 41 (2010) 888–901

Fig. 17. Detail of section M2, center.

Fig. 18. Detail of section M2, far right.

Fig. 19. Detail of section M1, far right.

From these tests it can be concluded that 6 out of 7 notched 5. Conclusions

specimens tested fail away from the open hole, and that no direct
correlation can be extracted between this behavior and the pres- A new material form is being considered for aircraft primary
ence of significant hot spots. Furthermore, during loading damage structural applications, and consists of discontinuous carbon fi-
initiates away from the hole, and sometimes in apparently pristine ber/epoxy random mat of chopped unidirectional prepreg tape.
regions. On the other hand, there appear to be no signs of damage Commercial applications for this type of material form already ex-
initiation and propagation around the hole, typical of continuous ist, such as Hexcel HexMCÒ. Non-destructive inspection of these
fiber laminates. materials is difficult due to the heterogeneity of the chip distribu-
 Author's personal copy

P. Feraboli et al. / Composites: Part A 41 (2010) 888–901 899

Fig. 20. Specimens #18–21, containing no evident hot spots, are loaded to failure in the presence of an open hole, and in 75% of cases lead to failure away from the hole.

Load increase

Fig. 21. Specimen #22 with an open hole undergoing progressive loading and ultrasonic imaging up to ultimate failure.

tion and orientation, which results in several regions of signal determine the type of the defect from non-destructive inspection
attenuation. A series of focused ultrasonic C-scan inspections, ver- alone. A series of tensile tests on pristine specimens, both unnot-
ified by the use of targeted microscopic analysis, have been used to ched and containing an open hole, are performed to measure the
characterize these regions of signal attenuations, referred to as growth of these hot spots up to ultimate failure. It is found that
‘‘hot-spots”. Previous studies by the authors have shown that these failure is likely not to occur in proximity of these hot spots, inde-
materials exhibit a particular notch-insensitive behavior, which is pendently from their size and location. For the notched specimens,
thought to be associated to the heterogeneous micro-structure of in the vast majority of cases failure occurs away from the open hole
the material. It is found that these hot spots correspond to real de- and in most cases it precipitates in regions of apparently pristine
fects, such as macro-voids (resulting from entrapped air in resin- micro-structure, that is away from evident hot spots. In conclusion,
starved regions), swirls (regions of fiber kinking resulting from it appears that for this class of materials conventional ultrasonic
high flow), or resin-rich pockets. There appears to be no way to inspection and defect classification for part acceptance may not
 Author's personal copy

900 P. Feraboli et al. / Composites: Part A 41 (2010) 888–901

Load increase

Fig. 22. Specimen #23 with an open hole undergoing progressive loading and ultrasonic imaging up to ultimate failure.

Load increase

Fig. 23. Specimen #24 with an open hole undergoing progressive loading and ultrasonic imaging up to ultimate failure.

be adequate, and it could result in an unacceptable high rejection versity) is gratefully acknowledged for his insightful comments.
rate, without at the same time providing increased confidence in The experiments were carried out in the Automobili Lamborghini
part integrity and performance. Advanced Composite Structures Laboratory (ACSL) of the Depart-
ment of Aeronautics and Astronautics at the University of Wash-
ington in Seattle, WA.
Acknowledgments

The authors wish to acknowledge Dr. Al Miller (Director, Boeing References

787 Technology Integration) for funding this research, and Dr. John
Halpin (JCH Consultants) and Dr. Larry Ilcewicz (Federal Aviation [1] Halpin JC. Stiffness and expansion estimates for oriented short fiber
composites. Polym Eng Sci 1969;3:732.
Administration) for sharing their technical opinions during the re- [2] Halpin JC, Pagano NJ. The laminate approx. for randomly oriented short fiber
search. Luciano Demasi (Assistant Professor at San Diego State Uni- composites. Polym Eng Sci 1969;3:720.
 Author's personal copy

P. Feraboli et al. / Composites: Part A 41 (2010) 888–901 901

[3] Kardos JL, Michno MJ, Duffy TA. Investigation of high performance short fiber [7] Feraboli P, Peitso E, Cleveland T, Stickler PB, Halpin JC. Notched behavior of
reinforced plastics. Final Report, Naval Air Systems Command, No. N00019-73- prepreg-based discontinuous carbon fiber/epoxy systems. Composites: Part A
C-0358; 1974. 2009;40(3):289–99.
[4] Boeing 787 features composite window frames, Reinforced plastics, [8] Feraboli P, Peitso E, Cleveland T, Stickler PB. Modulus measurement for
Application News, 2007;51(3):4 [March]. prepreg-based discontinuous carbon fiber/epoxy systems. J Compos Mater
[5] Porter J. Moving closer to the goal of cost effective complex geometry carbon 2009;43(19):947–1965.
composite parts, HPC4HPC special session. In: Proceedings of the 19th ASC [9] ASTM D3171-09. Standard test methods for constituent content of composite
technical conference, Atlanta, GA, September 2004. materials, vol. 15.03. West Conshohocken (PA): ASTM International; 2008.
[6] Feraboli P, Peitso E, Deleo F, Cleveland T, Stickler PB. Characterization of [10] Boeing standard test method for unnotched tension, D6-83079-61. The Boeing
prepreg-based discontinuous carbon fiber/epoxy systems: Part I. J Reinf Plast Co.; 2008.
Compos 2009;28(10):1191–214.