Composites Science and Technology 69 (2009) 1599–1606

Contents lists available at ScienceDirect

Composites Science and Technology

journal homepage: www.elsevier.com/locate/compscitech

Failure detection and monitoring in polymer matrix composites subjected

to static and dynamic loads using carbon nanotube networks
M. Nofar, S.V. Hoa *, M.D. Pugh
Concordia Center for Composites, Department of Mechanical and Industrial Engineering, Concordia University, Center for Research in Polymers and Composites (CREPEC),
Montreal, Quebec, Canada H3G 2W1

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

Article history: In this work, multiwall carbon nanotubes (MWCNTs) have been used as a network of sensors to predict
Received 28 June 2008 the failure region and to monitor the degradation of mechanical properties in laminated composites sub-
Received in revised form 26 February 2009 jected to tensile and cyclic fatigue loadings. This is achieved by measuring the electrical resistance change
Accepted 4 March 2009
in the semi-conductive MWCNT-fiber glass–epoxy polymer matrix composites. By partitioning the ten-
Available online 21 March 2009
sile and fatigue samples with electrically conductive probes, it is shown that with both increasing tensile
load and number of cycles different resistance changes are detected in different regions and failure hap-
Keywords:
pens in the part in which higher resistance change was detected. In cyclic loading, when compared to
A. Polymer composites
A. Carbon nanotubes
strain gauge readings, resistance change measurements show more sensitivity in identifying the crack
B. Electrical properties location, which gives this technique a good potential for monitoring damage during fatigue.
Sensors Ó 2009 Elsevier Ltd. All rights reserved.
A. Nanocomposites
A. Smart materials
B. Fatigue

1. Introduction tive fibers such as carbon fibers, using a dispersed conductive

Carbon nanotubes (CNT) are known as the stiffest and strongest
materials in the world [1–3]. They were discovered in the early
1990s and researchers started to focus on these new materials
due to their considerable physical and mechanical properties. Their
remarkable mechanical properties such as high stiffness and
strength, exceptional resilience, low density, fiber-like structure
with high aspect ratio (length/diameter), as well as high electrical
and thermal conductivity render carbon nanotubes (CNTs) poten-
tial as nanoscale reinforcement to achieve improved electrical
properties [4–9]. Recently, researchers have utilized them as strain
sensors by embedding them in polymer matrix composites (PMCs)
and monitoring damage and subsequently failure, by direct mea-
surement of current in the composite [10–13]. Damage monitoring
using electrical techniques has been of interest to researchers for
some time. It began with the use of carbon fibers as the conductive
reinforcement, so that fracture of the fibers would result in a
change of electrical resistance [14–16]. Since matrix-dominated
fracture mechanisms can not be monitored utilizing these conduc-
tive fibers and this also places a restriction on only using conduc-
* Corresponding author. Tel.: +1 514 848 2424; fax: +1 514 848 3175.
E-mail addresses: m_nofar@encs.concordia.ca (M. Nofar), hoasuon@alcor.con
cordia.ca, hoasuon@vax2.concordia.ca (S.V. Hoa), pugh@encs.concordia.ca (M.D.
Pugh).
reinforcement in the matrix has become attractive for damage
monitoring. The high aspect ratio of carbon nanotubes has been
considered as crucial to the formation of a conductive percolating
network through the polymer matrix composite at relatively low
CNT concentrations: such a network is highly sensitive to the onset
of damage in matrix-dominated fractures [10]. This exceptional
sensitivity provides an increase in electrical resistivity along
embedded carbon nanotubes in PMCs under even a low mechani-
cal load [17]. In comparison with resistance-type strain gauges,
which provide a conventional way for measuring strain, using con-
ductive CNT may not have the same limitations, such as measuring
strain at only specific locations and a lack of versatility and flexibil-
ity [18]. In Thostenson and Chou’s study, [10] unidirectional glass
fiber–epoxy composites with dispersed multiwall carbon nano-
tubes within the epoxy have been utilized to evaluate the damage
and percolation threshold in tensile and flexural tests as an in situ
sensor.
When considering non-static loads such as fatigue, which is one
of the most crucial factors for failure in many structural compo-
nents, the long term durability of polymer matrix composites is
highly dependent on the polymer matrix and the fiber–matrix
interface [19]. In other words, fatigue life and damage tolerance
are strongly affected by matrix cracks, like micro cracks appearing
between fiber reinforcements, or ply delamination between layers
[10]. Carbon nanotubes as additives can play a significant role as

0266-3538/$ - see front matter Ó 2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compscitech.2009.03.010
1600 M. Nofar et al. / Composites Science and Technology 69 (2009) 1599–1606

distributed sensors to monitor damage and to determine the
extent and defect propagation created by cyclic loads [20,21]. In
another study, Böger et al. [22] worked on stress/strain and dam-
age monitoring of glass fiber-epoxy using carbon nanotubes and
carbon black during tensile, fatigue, and interlaminar shear
strength testing.
Damage monitoring and detection of failure location is further
explored in this study.
2. Experimental
2.1. Primary materials
Multiwall carbon nanotubes grown by the chemical vapor
deposition technique (more than 95% purity) have been used in
this work (Cheap Tubes). They have diameters in the order of
10–20 nm and lengths of 10–30 lm. The epoxy polymer and curing
agent are bisphenol-F epoxy, EPIKOTE Resin 862, and EPIKURE W
(an aromatic amine curing agent), respectively (Hexion specialty
chemicals). Bidirectional woven glass fibers have been utilized as
the mechanical reinforcement.
2.2. Fabrication of nanotube-fiber–epoxy composites
To untangle and disperse the nanotubes within the epoxy to
make a network of sensors, two methods were considered; acetone
sonication and calendering. A calendering machine (Three-Roll Mill
EXAKT) was used to produce high shear force mixing while passing
the resin/nanotubes mixture through the rolls. The minimum
agglomeration was seen after gradually reducing the gap setting
from 50 lm to 5 lm. 1 wt% of MWCNTs (this amount selected
due to conductivity achievements reported in [23]) was dispersed
within the epoxy resin. The resin was then heated up to 50 °C to
reduce the viscosity and the curing agent (in the ratio of
26.4:100) was added to the epoxy. After stirring for 5 min, the mix-
ture was heated up to 70 °C for 20 min in a vacuum oven in order
to degas and to remove the bubbles. Fig. 1 shows that calendaring
provides better dispersion than ultrasonication. Electrical conduc-
tivity values for nanotubes dispersed in epoxy using acetone soni-
cation and calendering methods were 7.8  10 8 and 4.8  10 5 S/
cm, respectively. These are average values over five samples each
with a standard deviation of 5%. Considering this result, the calen-
dering method was utilized for dispersion of the nanotubes.
The final resin obtained was applied to two layers of bidirec-
tional plain weave glass fabrics (18 oz/yd2) by hand lay-up fol-
lowed by 2 h of vacuum bagging. The fabricated polymer matrix
composite was cured in an oven for 6 h at 130 °C. The average fiber
volume fraction in the laminate is 53 wt%. Sand paper and epoxy
862 were used to make the end tabs.
2.3. Electrical measurements under mechanical loads
Samples were cut out of the cured composite plate, according to
ASTM D 3039 for mechanical testing. To improve gripping of the
end tabs of the samples, screen sandpaper was bonded onto the
tabs. Silver epoxy-based glue was used as the contact for the
conductive probes for electrical resistance measurement. Measure-
ments were made using a high resistance meter (Agilent machine
4339B). Tensile and fatigue tests were carried out on an MTS
100KN universal testing machine and at the same time, a constant
source voltage was applied to the samples and the electrical cur-
rent was measured in order to calculate the electrical resistance
change.
2.4. Tensile tests
Tensile tests were performed at a rate of 1.27 mm/min while
measuring the resistance change. For a few samples, two electrical
probes were bonded to the ends of the gage area for electrical mea-
surement as shown in Fig. 2. For a few other samples, two addi-
tional electrical probes were bonded on the sample surface at
equal distance to measure the resistance changes along the sample
in three different zones as shown in Fig. 3. Electrical resistance
change is defined as (Ri R0/R0)  100, where R0 is the electrical
resistance of the measured region before loading, and Ri is the
resistance while loading.
Fig. 2 shows load versus strain and resistance change versus
strain measurements. The load–strain curve increases linearly until
a point where a change of slope occurs at a strain of about 5% (cor-
responding to a load of about 2000–3000 N). This change of slope
happens at the same strain in the resistance–strain curve. This
Fig. 2. Change of resistance in tension for 1 wt% nanotube-glass fiber–epoxy
composite.

Fig. 1. Nanotubes dispersion within epoxy by means of (A) acetone sonication and (B) calendering.
 M. Nofar et al. / Composites Science and Technology 69 (2009) 1599–1606
Fig. 3. Change of resistance versus strain by multi point resistivity measurement
during tensile test failure has occurred between (a) points 2 and 3, and (b) points 1
and 2.
point of slope change can be called the elastic limit of the sample. A
clear sound was heard when the load went past the elastic limit.
The resistance increase may be related to the formation of some
damage in the sample that will subsequently cause failure.
2.4.1. Resistance change at different regions in the gage area
Exploiting the fact that carbon nanotube network can provide
sensing at every point in the sample, it was of interest to examine
whether different regions of the gage area of the sample would
exhibit the same strain. In order to do this, two additional electrical
probes were bonded to the sample to provide three regions for
measuring electrical resistance. Two samples were tested. Fig. 3
shows the load–strain curves and resistance–strain curves. It can
be seen that different regions in the gage area show different resis-
tance changes. The change in resistance also corresponds to the
location of failure. For sample 1 (Fig. 3a), failure occurs in the
region between points 2 and 3 and the change in resistance is this
region is more than that in the other two regions, with sudden
large increase when the load is more than about 6500 N. The same
behavior is observed for sample 2, except that failure occurs
between points 1 and 2 for sample 2. Note also that the slope in
the load–strain curve changes (decreases) at a strain of about
0.5% and at the same time as the rate of resistance change starts
to increase. This reinforces the observation made in the results
shown in Fig. 2.
2.5. Fatigue test
2.5.1. Resistance increase with number of cycles
To explore the potential of a nanotube network to monitor dam-
age accumulation in a fatigue test, different tension–tension fati-
gue loads were applied. The minimum applied load was 150 N in
all cases and a range of maximum loads was used (1500, 2500,
3500, 4000, and 6000 N). Fig. 4 shows the resistance changes with-
in 500 load cycles. A sharp increase in resistance is observed within
the first several cycles and afterwards the resistance increases at a
lower rate. Also, the resistance change depends on the magnitude
1601
Fig. 4. Change of resistance versus cycles with maximum applied loads of 1500,
2500, 3500, 4000, and 6000 N within 500 cycles.
of the maximum load. The magnitude of the initial increase of
resistance is significantly smaller for maximum loads below the
elastic limit. A clear sound was emitted from samples that were
submitted to maximum loads of 3500, 4000, and 6000 N when
the load went past about 2000–3000 N.
2.5.2. Residual resistance change and residual strains
In Figs. 2 and 3, it was observed that there exists an elastic limit
at about 2000–3000 N. It would be of interest to examine the resid-
ual strain after fatigue loading at maximum loads below and above
this elastic limit. Fatigue tests were conducted up to 100 cycles
from a minimum load of 150 N to six different maximum loads
(500, 1500, 2500, 3500, 4500, and 6000 N). After every four cycles,
the sample was unloaded and residual resistance and strain were
measured. Electrical resistance change is defined as (Ri R0/
R0)  100, where R0 is the electrical resistance of the measured
region before loading, and Ri is the resistance when the sample is
unloaded.
Fig. 5 shows the residual measurements at six different maxi-
mum loads. For maximum loads of 1500 and 500 N which are
below the elastic limit range, no residual resistance change and
residual strain were observed. By increasing the maximum load
above 2500 N, the residual resistance change and strain become
apparent and increase with increasing cycles. These results confirm
the significance of the elastic limit.
It is also observed in Fig. 5 that at loads above the elastic limit,
the residual change in resistance is more sensitive to the load level
than is the residual strain. The resistance changes for loads of
3500 N and 4500 N are much higher than the change for a load
of 2500 N while the residual strains at loads of 3500 N and
4500 N are about the same as the residual strains at lower loads.
Only when the load becomes very high (6000 N) does the residual
strain shows significant increase. This again shows the sensitivity
of the carbon nanotube network in providing indication for the
state of deformation in the sample.
In order to see whether the increase in residual change in resis-
tance (while there is no significant increase in residual strain) cor-
responds to any damage in the material, samples that were tested
to a maximum of 3500 N were cut along the longitudinal direction.
The sections were polished and observed under a microscope.
Fig. 6 shows two of the sections. It can be seen that matrix cracks
occur in these samples. Examination of Fig. 5 shows that the mea-
surement of the residual change in resistance is more sensitive to
the effect of these cracks than the measurement of residual strain
change. The reason for this is due to the use of fabrics. Even though
there are matrix cracks in one layer, fibers in the other direction
1602 M. Nofar et al. / Composites Science and Technology 69 (2009) 1599–1606

Fig. 5. (a) Residual change of resistance and (b) residual strain using strain gauge measured for six maximum loads in the first 100 cycles.

Fig. 6. Micrographs of sections of samples that were tested to a maximum of 3500 N.

still carry the load and the matrix damage is not translated into sig-
nificant increase in strain.
2.5.3. Monitoring damage in fatigue testing
The observed sensitivity of the carbon nanotube network in indi-
cating the state of deformation in the composite samples gave rise
to the motivation to investigate its ability to monitor damage. For this
study, three electrical probes were bonded to the sample providing
two regions of equal dimensions (each region about 3.75 cm long)
in the gage area (Fig. 7). Three types of samples were made and tested:
Type I: Five samples (two layers of fabrics) with two regions
with strain gauges placed in the center of each region.
Type II: Two samples (two layers of fabric) with two regions
with a deliberate cut made in the lower region of the cured sample.
The first strain gauge is attached at the center of the top region and
the other one in the bottom region, nearer the top region to allow
some distance from the defect.
Type III: Two samples with three layers of glass fiber rather than
two. The mid layer has a cut in the bottom region before curing.
After manufacturing of the composite, the defective part of the
sample was marked in order to identify the artificial crack made
in the mid layer. Strain gauges were positioned at the center of
each region. The strain gauge in the lower region is located about
2–3 mm above the crack location.
Maximum cyclic loads of 4500 and 6000 N were applied for
type I samples, 2000 and 3000 N for type II samples, and 6000 N
for type III samples. After every four cycles of loading, the sample
was unloaded and electrical resistance and strain values were mea-
sured. After a number of cycles, a tensile test was conducted to find
the failure location.
 M. Nofar et al. / Composites Science and Technology 69 (2009) 1599–1606
Fig. 7. Three types of samples for damage monitoring in two regions in fatigue test.
2.5.3.1. Fatigue testing of type I samples. In type I, five samples were
tested. Fig. 8a shows the responses for sample 3 with a maximum
load of 4500 N. After conducting the tensile test, failure occurred
between points 1 and 2 which was very close to the upper strain
gauge (almost 1–2 mm). In this case both residual strain and resid-
ual change in resistance show higher values for the region between
points 1 and 2 (as compared to the region between points 3 and 4),
which corresponds to the location of the crack.
Fig. 8b shows the response for sample 5, with a maximum load
of 6000 N. The crack occurs in the region between points 2 and 3. In
Fig. 8. Residual change of resistivity and residual strain in two regions, after 100 fatigue cycles with maximum load of 4500 N (type I, sample 3). Residual change of resistance
and residual strain in two regions, after 100 fatigue cycles with maximum load of 6000 N (type I, sample 5).
1603
this case, the strain gage values in the two regions do not differ,
while the residual change in resistance in regions 2 and 3 is much
higher than that in regions 1 and 2. This result shows that the
residual change in resistance corresponds better with the crack
location. The results from the other three samples of type I also
shows similar pattern. Strain gage values sometime correspond
to the location of the crack, and sometimes not. On the other hand,
residual change in resistance always corresponds to the location of
the crack.
Photograph of a few broken samples are shown in Fig. 9.
2.5.3.2. Fatigue testing of type II samples. Similar tests were carried
out on two samples of type II. Since a cut has been made in the
lower region between points 2 and 3, to decrease the effect of local
deformation around that cut, a strain gauge was placed at a dis-
tance of 2.5 cm from point 3 (Fig. 10a and b). Fatigue testing was
implemented with maximum cyclic loads of 2000 and 3000 N,
knowing that the failure would occur in the defective part. As
shown in Fig. 10a and b, the resistance measurements show higher
rate of change between points 2 and 3 in which the failure initia-
tion takes place.
2.5.3.3. Fatigue testing of type III samples. Two samples of type III as
depicted in Figs. 11a and b were fatigued with a maximum load of
6000 N. The residual strains between points 2 and 3 increase more
than that between points 1 and 2. The resistance increases show
the same behavior.
1604 M. Nofar et al. / Composites Science and Technology 69 (2009) 1599–1606

Fig. 9. Photograph of two broken samples.

Fig. 10. (a) Residual change of resistivity and residual strain in two regions, after 100 fatigue cycles with maximum load of 2000 N (type II, sample 1). (b) Residual change of
resistivity and residual strain in two regions, after 100 fatigue cycles with maximum load of 3000 N (type II, sample 2).

3. Discussion
Two important conclusions can be obtained from the experimen-
tal results. First is the detection of an elastic limit due to the change in
slope of the electrical resistance versus strain curve, and second is
the better sensitivity of the carbon nanotube network in the moni-
toring of damage detection as compared to strain gages.
The detection of the elastic limit is supported not only by the
change in slope of the electrical resistance versus strain curve (Figs.
2 and 3) but also by the absence of residual change in resistance for
fatigue loading up to maximum loads that are lower than the elas-
tic limit (Fig. 5).
The better sensitivity of the carbon nanotube network in detect-
ing damage is supported by the more sensitive residual change in
resistance (Fig. 5) coupled with the presence of matrix cracks for
loads where there is large change in residual resistance and little
change in residual strain (Fig. 6). This is further supported by fati-
gue results done on starting samples without cracks (Fig. 8a and b)
and starting samples with cracks (Figs. 10 and 11). The better sen-
sitivity of the carbon nanotube network as compared to strain
gages can be explained by the fact that carbon nanotubes are
spread throughout the matrix in the composites, and most of the
initial cracks and delaminations take place within the matrix mate-
rial. Strain gages can sometimes reveal the presence of cracks and
sometimes not. This depends on the orientation of the cracks as
compared to the orientation of the strain gage. If the crack happens
to be delamination of fibers parallel to the strain gage length direc-
tion, then the strain gage may not pick it up. If the crack orientation
 M. Nofar et al. / Composites Science and Technology 69 (2009) 1599–1606
Fig. 11. (a) Residual change of resistance and residual strain in two regions, after 150 fatigue cycles with maximum load of 6000 N (type III, sample 1). (b) Residual change of
resistivity and residual strain in two regions, after 150 fatigue cycles with maximum load of 6000 N (type III, sample 2).
is normal to the direction of the strain gage, then it is more prob-
able that the strain gage will feel the crack. In addition, whether
the strain gage will feel the presence of the crack depends on the
distance from the gage to the crack and on the ductility of the
material. The stress path around a crack may go around the gauge
if the strain gauge is located too close to the crack. On the other
hand, since the nanotube networks are connected all over the
sample, the occurrence of any defect or damage can cause discon-
figuration of the nanotube network. This in turn produces an in-
crease in resistance along the sample, regardless of the location of
failure.
The above points show that carbon nanotubes have the poten-
tial to monitor the strength degradation during static and dynamic
loading and to predict failure before it happens. This self-sensing
method using nanotubes as a network of sensors can also be pro-
posed for failure location prediction in polymer matrix composites.
4. Conclusions
Towards the development of a new method to monitor failure
under static and dynamic loads, multiwall carbon nanotubes were
incorporated in an epoxy matrix to form a network of sensors.
During static tensile loading, location of the failure could be deter-
mined based on higher electrical resistance change in one zone
compared to the other zones. Also in different zones, by applying
cyclic loads and comparing the resistance change with strain gauge
measurements, it was seen that the electrical resistance measure-
ments show more consistency in correlation with failure location.
Failure location can be predicted by higher increase of resistance
change in a region compared to other regions of the sample. The
formation of matrix cracks causes disconfiguration of the nanotube
network and subsequent reduction of the electrical current flow
through them. These results show that nanotube networks have
the potential to be used for monitoring the integrity of composite
laminates during both static and cyclic loads.
1605
Acknowledgments
The authors gratefully acknowledge financial supports of Natu-
ral Sciences and Engineering Research Council of Canada (NSERC),
Center for Research in Polymers and Composites (CREPEC) and
Concordia University.
References
[1] Wong EW, Sheehan PE, Lieber CM. Nanobeam mechanics: elasticity, strength,
and toughness of nanorods and nanotubes. Science 1997;277:1971–5.
[2] Yu M, Lourie O, Dyer MJ, Kelly TF, Ruoff RS. Strength and breaking mechanism
of multiwalled carbon nanotubes under tensile load. Science
2000;287:637–40.
[3] Yu MF, Files BS, Arepalli S, Ruoff RS. Tensile loading of ropes of single wall
carbon nanotubes and their mechanical properties. Phys Rev Lett
2000;84(24):5552–5.
[4] Ruoff RS, Lorents DC. Mechanical and thermal-properties of carbon nanotubes.
Carbon 1995;33(7):925–30.
[5] Mintmire JW, White CT. Electronic and structural-properties of carbon
nanotubes. Carbon 1995;33(7):893–902.
[6] Wildoer JWG, Venema LC, Rinzler AG, Smalley RE, Dekker C. Electronic
structure of atomically resolved carbon nanotubes. Nature 1998;391:59–62.
[7] Li C, Chou TW. A structural mechanics approach for the analysis of carbon
nanotubes. Int J Solids Struct 2003;40(10):2487–99.
[8] Falvo MR, Clary GJ, Taylor II RM, Chi V, Brooks FP, Washburn S. Bending and
buckling of carbon nanotubes under large strain. Nature 1997;389:582–4.
[9] Iijima S, Brabec C, Mati A, Bernholc J. Structural flexibility of carbon nanotubes.
J Chem Phys 1996;104(5):2089–92.
[10] Thostenson ET, Chou TW. Carbon nanotube networks: sensing of distributed
strain and damage for life prediction and self healing. Adv Mater
2006;18:2837–41.
[11] Li Z, Dharap P, Nagarajaiah S, Barrera EV, Kim JD. Carbon nanotube film
sensors. Adv Mater 2004;16(7):640–3.
[12] Dharap P, Li Z, Nagarajaiah S, Barrera EV. Nanotube film based on single-wall
carbon nanotubes for strain sensing. Nanotechnology 2004;15:379–82.
[13] Zhang W, Suhr J, Koratkar N. Carbon nanotube/polycarbonate composites as
multifunctional strain sensors. J Nanosci Nanotechnol 2006;6:960–4.
[14] Weber I, Schwartz P. Monitoring bending fatigue in carbon-fibre/epoxy
composite strands: a comparison between mechanical and resistance
techniques. Compos Sci Technol 2001;61:849–53.
[15] Kupke M, Schulte K, Schuler R. Non-destructive testing of FRP by dc and ac
electrical methods. Compos Sci Technol 2001;61:837–47.
1606 M. Nofar et al. / Composites Science and Technology 69 (2009) 1599–1606
[16] Schueler R, Joshi SP, Schulte K. Damage detection in CFRP by electrical
conductivity mapping. Compos Sci Technol 2001;61:921–30.
[17] Parka JM, Kima DS, Leeb JR, Kim TW. Nondestructive damage sensitivity and
reinforcing effect of carbon nanotube/epoxy composites using electro-
micromechanical technique. Mater Sci Eng C 2003;23:971–5.
[18] <http://www.vishay.com>.
[19] Chou TW. Microstructural design of fiber composites. Cambridge
(UK): Cambridge University Press; 1992.
[20] Zhang W, Sakalkar V, Koratkar N. In situ health monitoring and repair in
composites using carbon nanotube additives. Appl Phys Lett 2007;91(13):1–3.
[21] Giang TP, Park YB, Liang Z, Zhang C, Wang B. Processing and modeling of
conductive thermoplastic/carbon nanotube films for strain sensing.
Composites Part B: Eng 2008;39:209–16.
[22] Böger L, Wichmann MHG, Meyer LO, Schulte K. Load and health
monitoring in glass fibre reinforced composites with an electrically
conductive nanocomposite epoxy matrix. Compos Sci Technol 2008;68:
1886–94.
[23] Thostenson ET, Chou TW. Processing-structure-multi-functional property
relationship in carbon nanotube/epoxy composites. Carbon 2006;44:
3022–9.