COMMUNICATION
DOI: 10.1002/adma.200600977
Carbon Nanotube Networks: Sensing of Distributed Strain and
Damage for Life Prediction and Self Healing**
By Erik T. Thostenson and Tsu-Wei Chou*
Increasingly, materials are being tailored to achieve multi-
functional properties where they can combine active, sensory,
adaptive, and autonomic capabilities. The expansion of pro-
cessing techniques that enable structuring materials at the
nanoscale combined with development of new methods for
analysis should enable optimization of material structure to
achieve systems that satisfy specific functional require-
ments.[1–3] With their well-known novel mechanical and elec-
trical properties, carbon nanotubes are inherently multifunc-
tional and can serve as a structural reinforcement as well as a
platform for sensing. In this work we demonstrate that
conducting carbon nanotube networks formed in an epoxy
polymer matrix can be utilized as highly sensitive sensors for
detecting the onset, nature, and evolution of damage in ad-
vanced polymer-based composites. Using direct-current mea-
surements the internal damage accumulation can be moni-
tored in situ. After the onset of damage and subsequent
reloading of the damaged structure there is a remarkable shift
in the sensing curve, indicating irreversible damage. These
results hold promise for evaluation of autonomic self-healing
approaches for polymers and development of enhanced life-
prediction methodologies.
Owing to their distinctive mechanical and physical proper-
ties, nanotubes offer unique potential for reinforcing poly-
mers. Considerable interest has focused on utilizing nanotubes
as passive reinforcement to tailor their mechanical,[4–7] electri-
cal,[8–11] and thermal properties.[12,13] Baughman et al.[14] first
reported the electromechanical actuation behavior of nano-
tubes. Inherent coupling of electrical and mechanical proper-
ties makes nanotubes excellent candidates for in situ sensors.
Recent reports have utilized nanotube-based materials as
electromechanical actuators[15,16] and in a variety of sensing
applications,[17] including mass sensors,[18] humidity sensors,[19]
and strain sensors.[20,21]
– sion of damage.
[*] Prof. T.-W. Chou, Prof. E. T. Thostenson
Department of Mechanical Engineering and
Center for Composite Materials
University of Delaware
Newark, DE 19716 (USA)
E-mail: chou@udel.edu
[**] This work is funded by the US Air Force Office of Scientific Research
(FA9550-06-1-0489, FA9550-04-1-0337—Dr. Byung-Lip Lee, Pro-
gram Director) as well as the National Science Foundation (NIRT
Program, Grant No. 0304506—Dr. Ken P. Chong, Program Direc-
tor). Supporting Information is available online from Wiley Inter-
Science or from the author.
Adv. Mater. 2006, 18, 2837–2841 © 2006 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
Long-term durability and performance of advanced fiber
composites are governed by properties of the polymer matrix
and the fiber/matrix interface.[22] In particular, fatigue life and
damage tolerance are strongly affected by matrix cracks oc-
curring between fibers as microcracks or between layers as
ply delamination. Although in-plane fracture behavior of
composites is dominated principally by breakage of load-car-
rying fibers, initiation of damage in the polymer matrix leads
to premature fracture and reduced durability. These matrix-
dominated damage mechanisms often occur at low strain or
because of out-of-plane loading. As a consequence, consider-
able interest has emerged in the development of self-healing
methodologies to minimize long-term effects of matrix crack-
ing.[23–26]
Utilizing electrical techniques has been established as a
noninvasive way to monitor damage in carbon-fiber-rein-
forced composites under static and dynamic loading condi-
tions.[27–30] Because carbon fibers are conductive, fracture of
fibers will result in changes in electrical resistance. This
approach does not give much insight into matrix-dominated
mechanisms of fracture that affect durability and is not applic-
able to composites where fibers are non-conducting (such as
glass or aramid fibers). Others have attempted to locally mod-
ify electrical properties of glass fiber yarn bundles through
coating with carbon powder and utilizing them for damage
sensing.[31] In their review, Fiedler et al.[32] first introduced the
concept of conductive modification with nanotubes as having
potential for damage sensing.
In the present work we process glass fiber–epoxy compos-
ites and utilize multiwalled nanotubes dispersed in the epoxy
phase as distributed sensors to evaluate the onset and evolu-
tion of damage in advanced fibrous composites. Here, we
show carbon nanotubes that form a conductive percolating
network throughout the polymer matrix are remarkably sensi-
tive to initial stages of matrix-dominated failure. Through
experiments designed to promote different failure modes we
demonstrate the possibility to identify the nature and progres-
If the high aspect ratio (length/diameter) of carbon nano-
tubes is preserved while processing a polymer nanocomposite,
the nanotubes can form a conductive percolating network
throughout the polymer matrix at relatively low concentra-
tions. In our ongoing research we have processed polymer
nanocomposites using a high shear stress field.[33] This tech-
nique maintains the relatively large aspect ratio of the nano-
tubes and results in percolation thresholds, where there is a
sharp decrease in the composite volume resistivity, at or be-
2837
COMMUNICATION

low 0.1 wt % carbon nanotubes, as shown in Figure 1a. A ten- of conventional fibers the nanotubes are able to penetrate
sile test performed on a nanotube/epoxy specimen (Fig. 1b) throughout and form a conductive percolating network in the
also shows a highly liner relationship between the specimen polymer matrix. This network formation in the polymer ma-
deformation and electrical resistance. trix offers significant potential to develop hierarchal sensing
Owing to the difference in reinforcement scale between approaches for damage detection and health monitoring.
conventional micrometer-sized fiber reinforcement and car- Electrical percolation with multiwalled carbon nanotubes at
bon nanotubes with nanometer-sized diameters, it is possible extremely low volume fractions enables the creation of in situ
to have carbon nanotube reinforcement in the matrix-rich sensors that are minimally invasive and not likely to substan-
areas between fibers in an individual bundle as well as be- tially alter the in-plane mechanical properties of the fiber
tween adjacent plies. Recent research has shown that the op- composite.
portunity exists for tailoring the properties of fiber composites In order to evaluate and assess the formation of nanotube
through nanoscale hybridization, where nanoscale reinforce- networks for in situ sensing we processed nanotube–glass fi-
ment can be included to improve the out-of-plane composite ber–epoxy composites and designed specific mechanical tests
properties or to stiffen the fiber/matrix interface.[34–36] For to promote distinct failure modes. Key experiments in tension
mechanical reinforcement, the effectiveness of multiwalled are to evaluate interlaminar delamination in unidirectional
carbon nanotubes is limited by the intratube van der Waals composites and transverse microcracking in cross-ply lami-
interaction that reduces the effective stiffness of the nanotube nates. When subject to bending, unidirectional specimens
in the polymer matrix.[6] In their recent study Gojny et al.[37] were tested with different spans to promote different failure
concluded that multiwalled carbon nanotubes offer the high- modes. In the current work, the specimen strain was not re-
est potential for enhancement of electrical conductivity. By corded using an external extensometer or bonded strain gages
first dispersing the nanotubes in the polymer matrix and then in order to avoid any influence on the resistance measure-
infiltrating the dispersed mixture through layers and bundles ments. The deformation data is instead presented in terms of
the crosshead displacement of the testing machine and is pro-
portional to strain.
18
a) 10
A five-ply unidirectional composite was fabricated where
16
10 the center ply of the laminate was cut in the middle of the
Ω cm)

14
specimen to promote ply delamination during tensile loading.
10
Volume Resistivity (Ω

The discontinuity at the center ply of the laminate results in

12
10 the accumulation of shear stresses at the ends of the ply and
10
10
these shear stresses initiate delamination of the center and ad-
jacent plies. Figure 2a shows the results of the tensile test. The
8
10 specimen resistance increases linearly with initial deformation
10
6 and is consistent with our earlier observation of linear in-
4
crease in resistance with deformation in nanotube–epoxy
10
specimens. A sharp increase in resistance occurs when the ply
100 delamination is initiated. As the delamination grows with in-
0 0.2 0.4 0.6 0.8 1
creasing load there is a large increase in resistance marked
Nanotube Weight (%) primarily by a progressive increase in the slope of the resis-
tance curve with extension of the ply delamination.
b) 500 10
The influence of transverse microcrack development in
plies oriented normal to the direction of loading was investi-
Resistance Change (%)

400 8 gated using a [0/90]s cross-ply specimen with plies oriented

along the loading axis (0°) on the outside of the laminate and
plies oriented at 90° in the center. It is widely known in cross-
Load (N)

300 6
ply laminates loaded in tension that failure initiates in the 90°
plies by the formation of microcracks that are oriented nor-
200 4
mal to the direction of applied load. As the laminate is further
strained more cracks are formed until the ply becomes satu-
100 2
rated with a regularly spaced array of cracks and the crack
density is related to laminate configuration.[22]
0 0 As observed in the unidirectional specimen there is a linear
0 0.1 0.2 0.3 0.4 0.5 0.6
increase in resistance with initial loading in the cross-ply spec-
Displacement (mm) imens (Fig. 2b). Upon the initiation of microcracking in the
Figure 1. a) Electrical percolation behavior in nanotube–epoxy compos-
90° plies there is a sharp change in the resistance. In the pro-
ites and b) resistance change with deformation for a 0.5 wt % nanotube– gression from the first initiation of cracking to ultimate failure
epoxy composite loaded in tension. of the composite laminate the resistance changes are marked

2838 www.advmat.de © 2006 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim Adv. Mater. 2006, 18, 2837–2841
 COMMUNICATION
a) 10000 400
the change in slope is small relative to the overall specimen
stiffness. Additional experiments were assessed to evaluate
350
the resistance-deformation behavior of pre- and post-damage

Resistance Change (%)

8000
300 initiation. Figure 2c shows the changes in resistance with de-
formation during loading of the composite laminate. The
250
Load (N)

6000 specimen was first loaded until the initial step increases re-
Delamination
Extension 200 lated to the initiation of microcracking was observed. Upon
4000 150 unloading, the resistance decreased to nearly the original val-
ue as the transverse cracks were closed by the stiff outer plies.
100
2000 Upon reloading, the specimen showed sharp increases in resis-
Delamination 50 tance at much lower levels of deformation corresponding to
Initiation reopening of the microcracks, indicative of permanent dam-
0 0
0 0.5 1 1.5 2 2.5 3 3.5 4 age to the composite. This approach could be a powerful tech-
Displacement (mm) nique to evaluate the effectiveness of self-healing methodolo-
gies. After initial cracking the residual stiffness of the
b) 4000 70
Damage composite is still quite high, and evaluation of self-healing
3500 Accumulation 60 strategies using residual stiffness measurements do not pro-
3000
Resistance Change (%) vide direct insight toward the effectiveness and extent of ma-
50
trix healing because the stiffness is fiber dominated. Monitor-
2500
Load (N)

40 ing the shift in the resistance curves on loading and unloading

2000 should directly correspond to accumulated damage. The spec-
30 imen is first loaded in tension to initiate damage and then
1500
20
unloaded to obtain the damaged resistance curve. After the
1000 material is allowed sufficient time to heal the resulting resis-
500 10 tance curve upon reloading should shift toward the undam-
aged resistance curve and would provide a direct assessment
0 0
0 0.25 0.5 0.75 1 1.25 1.5 of healing effectiveness.
Displacement (mm) For laminated composites the reinforcement is primarily
oriented in-plane. When subject to bending the layers of the
c) 20
composite transfer the load through shear stresses at the ply
interfaces. Composites are particularly susceptible to failure
Resistance Change (%)

15 Initial Loading at the ply interfaces because the polymer matrix must transfer
(undamaged) the entire shear load from layer to layer. When subject to out-
Second Loading
(damaged) of-plane loads, such as low velocity impact, the local shear
10 stress can result in delamination of the plies. In order to assess
the capability of nanotubes to sense through-thickness interla-
minar fracture, unidirectional specimens were tested in three-
5 point bending at varying spans. As the span is increased, the
moment is higher and will produce stresses high enough to
promote fiber fracture as in the initial deformation mecha-
0
0 0.2 0.4 0.6 0.8 1 1.2 nism. Specimens were tested at span-to-thickness ratios of
four and eight to promote the different types of failure. Fig-
Displacement (mm)
ure 3a and b shows the load versus midspan displacement re-
Figure 2. a) Load-displacement and resistance curves for the 0 ° speci- sults for the short and long spans, respectively. With initial
men with center ply cut to initiate delamination; b) load-displacement loading both show a slight decrease in the resistance as the
and resistance curves for the 0/90 specimen, c) resistance curves for ini- beam is deformed in both tension and compression. At the
tial loading (undamaged) and reloading (damaged) laminates.
point of failure for the beam with the short span, the resis-
tance increases by several orders of magnitude indicating that
by step increases corresponding to the accumulation of micro- the specimen has delaminated completely. Viewing the side of
cracks and linear increases with deformation between the step the as-tested specimen, shown in Figure 3c, clearly reveals the
increases. By comparing resistance curves in Figure 2a and b formation of an interlaminar crack. Subsequent reloading and
it is possible to identify the nature and progression of damage large resistance spikes are likely a consequence of interlami-
using nanotubes. nar friction and the establishment of new electrical contacts.
For both the 0 and 0/90 specimens there is a slight degrada- For the long-span specimen there are slight decreases in load
tion in the stiffness with increasing damage accumulation, but that correspond to a flattening of the resistance curve. These

Adv. Mater. 2006, 18, 2837–2841 © 2006 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim www.advmat.de 2839
COMMUNICATION

a) 9
b)
600 10 350 35
Interlaminar Fiber Breakage
8 Delamination
Failure 10 300 and Crushing 30
500 t Extension

Resistance Change (%)

Resistance Change (%)

7 t
10 250 25
4t
400
8t

Load (N)
Load (N)

6 20
10 200
300 5
10 15
150
200 4
10 10
100

1000 5
100 50
0
100
0 0
0 0.2 0.4 0.6 0.8 1 1.2 0 0.5 1 1.5 2 2.5

Displacement (mm) Displacement (mm)

c) d)

Figure 3. Flexural test results showing a) span-to-thickness ratio of four to promote interlaminar failure and b) span-to-thickness ratio of eight to pro-
mote fiber failure. Optical microscopic images of a) interlaminar fracture (span/thickness = 4) and b) fiber and matrix damage under load support
(span/thickness = 8).

load drops are likely due to fiber breakage and crushing under
the load nose. The side of the specimen, Figure 3d, shows fi-
ber and matrix damage directly below the midspan load point.
Subsequent sharp decreases in load are accompanied by step
increases in resistance and likely correspond to matrix crack-
ing and the extension of local delamination. Unlike the speci-
men that fails in interlaminar shear, where the resistance
changes several orders of magnitude corresponding to com-
plete ply separation, the more incremental changes in resis-
tance correspond to damage accumulation.
In summary, in situ monitoring using carbon nanotubes of-
fers potential as both a laboratory tool for evaluating damage
progression during testing as well as for in-service health
monitoring. The exploration of strategies for sensor config-
uration and tailoring nanotube concentration will help shed
light on the progression and characteristic of damage states.
We have shown that conductive percolating nanotube net-
works in traditional fiber composites can accurately detect the
onset, nature, and progression of damage. The sensitivity of
the technique for damage sensing may have broad applica-
tions including the assessment of self-healing strategies.
Experimental
Multiwalled carbon nanotubes grown using chemical vapor deposi-
tion techniques with purity higher than 95 % (Iljin Nanotech) were
used to create the percolating network of sensors. Scanning electron
microscopy (SEM) images of as-grown carbon nanotubes show they
are highly entangled and have diameters on the order of 20–30 nm
and lengths of several micrometers. The epoxy polymer matrix used is
a bisphenol-F epoxy resin cured with an aromatic amine curing agent
(EPIKOTE Resin 862/EPIKURE W Curing Agent; Hexion Specialty
Chemicals).
To fabricate the nanotube–epoxy–fiber composites the carbon
nanotubes were first dispersed in the epoxy resin using a calendering
approach. A laboratory-scale three-roll mill (EXAKT-80E; EXAKT
Technologies) was utilized to impart high shear mixing to the nano-
tube–epoxy mixture to untangle and disperse the nanotubes. Nano-
tubes were weighed out for a final concentration of 0.5 wt %, added
to the epoxy resin by hand mixing and then processed using a three-
roll mill. The mixture was processed by passing through the mill at
progressively decreasing gap settings down to 5 lm. The evolution of
nanoscale composite structure during the process was evaluated using
transmission electron microscopy (JEOL 2000FX; 200 kV) to ensure
a high degree of dispersion.
Vacuum-assisted resin transfer molding was used to fabricate the fi-
ber–epoxy composites with embedded carbon nanotubes. Unidirec-

2840 www.advmat.de © 2006 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim Adv. Mater. 2006, 18, 2837–2841

tional nonwoven glass fiber mats were laid up and stacked in pre-
ferred orientations. For cross-ply composites [0/90]s, the laminate con-
sisted of four layers with two layers on the top and bottom with a
zero-degree orientation to the load direction and two layers in the
middle orthogonal to the top and bottom plies. Unidirectional com-
posites were five layers of fabric oriented in the same direction [0]5
where the center ply was cut to initiate delamination during tensile
loading. Composites tested in flexure were eight layers of unidirec-
tional fabric [0]8. The lay up was then compacted under vacuum using
a vacuum bag. After dispersing nanotubes in the epoxy resin it was
heated to 50 °C to reduce the viscosity and then mixed with the curing
agent at a ratio of 26.4:100 curing agent to epoxy. The mixture was
mechanically stirred and degassed in a vacuum oven for 15 min at
50 °C. Composites were fabricated by drawing the resin mixture
through the glass fibers under vacuum and then curing for 6 h at
130 °C.
Test specimens were machined from the cured composite laminates
to the desired dimensions using a slot grinder. End tabs were bonded
to tensile specimens to minimize the stress concentration caused by
the grips of the mechanical testing machine. The material used for the
end tabs was a non-conducting glass–epoxy composite to electrically
insulate the conductive composite laminate from the metallic speci-
men grips. Nanotube–epoxy tensile bars were electrically insulated by
curing a film of neat epoxy in the grip area and the three-point bend
specimens were electrically insulated from the load supports using
electrical tape. Mechanical testing in tension and flexure was per-
formed using a screw-driven load frame (Instron 5587), and specimen
deformation is expressed in terms of machine crosshead displace-
ment.
Electrodes were applied to the composites using conductive silver
paint and changes in resistance were measured using a highly sensitive
voltage–current meter (Keithley 6430). For the unidirectional and
nanocomposite tensile specimens the electrodes were applied to the
specimen surfaces just below the end tabs and on opposing sides.
Electrodes for the cross-ply specimens were applied directly to the
specimen ends, and the electrodes for flexural testing were applied to
the specimen tensile and compressive sides. A computer interface was
used to record the transient changes in resistance during mechanical
testing. Because the mechanical tests were performed at fixed cross-
head-displacement rates, the transient resistance data could be di-
rectly related to test displacement. A constant source voltage was
applied to the specimens during testing and the current was measured
to calculate the resistance changes. The source meter was also used to
measure the DC volume resistivity of the nanotube–epoxy compos-
ites.
Received: May 5, 2006
Revised: July 10, 2006
Published online: October 2, 2006
– Appl. Phys. 2002, 91, 6034.
[1] E. T. Thostenson, C. Y. Li, T. W. Chou, Compos. Sci. Technol. 2005,
65, 491.
[2] E. T. Thostenson, Z. F. Ren, T. W. Chou, Compos. Sci. Technol.
2001, 61, 1899.
[3] Y. Dzenis, Science 2004, 304, 1917.
[4] J. Suhr, N. Koratkar, P. Keblinski, P. Ajayan, Nat. Mater. 2005, 4, 134.
______________________
Adv. Mater. 2006, 18, 2837–2841 © 2006 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim
COMMUNICATION
[5] F. H. Gojny, M. H. G. Wichmann, U. Kopke, B. Fiedler, K. Schulte,
Compos. Sci. Technol. 2004, 64, 2363.
[6] E. T. Thostenson, T. W. Chou, J. Phys. D: Appl. Phys. 2003, 36, 573.
[7] E. T. Thostenson, T. W. Chou, J. Phys. D: Appl. Phys. 2002, 35, L77.
[8] F. M. Du, J. E. Fischer, K. I. Winey, Phys. Rev. B 2005, 72, 121 404.
[9] B. Vigolo, C. Coulon, M. Maugey, C. Zakri, P. Poulin, Science 2005,
309, 920.
[10] L. Wang, Z. M. Dang, Appl. Phys. Lett. 2005, 87, 042 903.
[11] J. K. W. Sandler, J. E. Kirk, I. A. Kinloch, M. S. P. Shaffer, A. H.
Windle, Polymer 2003, 44, 5893.
[12] H. Huang, C. H. Liu, Y. Wu, S. Fan, Adv. Mater. 2005, 17, 1652.
[13] M. J. Biercuk, M. C. Llaguno, M. Radosavljevic, J. K. Hyun, A. T.
Johnson, J. E. Fischer, Appl. Phys. Lett. 2002, 80, 2767.
[14] R. H. Baughman, C. X. Cui, A. A. Zakhidov, Z. Iqbal, J. N. Barisci,
G. M. Spinks, G. G. Wallace, A. Mazzoldi, D. De Rossi, A. G. Rin-
zler, O. Jaschinski, S. Roth, M. Kertesz, Science 1999, 284, 1340.
[15] S. V. Ahir, E. M. Terentjev, Nat. Mater. 2005, 4, 491.
[16] G. M. Spinks, V. Mottaghitalab, M. Bahrami-Saniani, P. G. Whitten,
G. G. Wallace, Adv. Mater. 2006, 18, 637.
[17] C. Y. Li, T. W. Chou, J. Intell. Mater. Syst. Struct. 2006, 17, 247.
[18] C. Y. Li, T. W. Chou, Appl. Phys. Lett. 2004, 84, 5246.
[19] P. G. Su, S. C. Huang, Sens. Actuators B 2006, 113, 142.
[20] P. Dharap, Z. L. Li, S. Nagarajaiah, E. V. Barrera, Nanotechnology
2004, 15, 379.
[21] J. Halary, P. Cookson, J. L. Stanford, P. A. Lovell, R. J. Young, Adv.
Eng. Mater. 2004, 6, 729.
[22] T. W. Chou, Microstructural Design of Fiber Composites, Cambridge
University Press, Cambridge, UK 1992.
[23] S. R. White, N. R. Sottos, P. H. Geubelle, J. S. Moore, M. R. Kessler,
S. R. Sriram, E. N. Brown, S. Viswanathan, Nature 2001, 409, 794.
[24] E. N. Brown, N. R. Sottos, S. R. White, Exp. Mech. 2002, 42, 372.
[25] E. N. Brown, S. R. White, N. R. Sottos, Compos. Sci. Technol. 2005,
65, 2466.
[26] E. N. Brown, S. R. White, N. R. Sottos, Compos. Sci. Technol. 2005,
65, 2474.
[27] K. Schulte, C. Baron, Compos. Sci. Technol. 1989, 36, 63.
[28] I. Weber, P. Schwartz, Compos. Sci. Technol. 2001, 61, 849.
[29] M. Kupke, K. Schulte, R. Schuler, Compos. Sci. Technol. 2001, 61,
837.
[30] R. Schueler, S. P. Joshi, K. Schulte, Compos. Sci. Technol. 2001, 61,
921.
[31] N. Muto, Y. Arai, S. G. Shin, H. Matsubara, H. Yanagida, M. Sugita,
T. Nakatsuji, Compos. Sci. Technol. 2001, 61, 875.
[32] B. Fiedler, F. H. Gojny, M. H. G. Wichmann, W. Bauhofer,
K. Schulte, Ann. Chim.- Sci. Mat. 2004, 29, 81.
[33] E. T. Thostenson, T. W. Chou, Carbon, doi: 10.1016/j.car-
bon.2006.05.014.
[34] E. T. Thostenson, W. Z. Li, D. Z. Wang, Z. F. Ren, T. W. Chou, J.
[35] F. H. Gojny, M. H. G. Wichmann, B. Fiedler, W. Bauhofer,
K. Schulte, Composites Part A 2005, 36, 1525.
[36] V. P. Veedu, A. Y. Cao, X. S. Li, K. G. Ma, C. Soldano, S. Kar, P. M.
Ajayan, M. N. Ghasemi-Nejhad, Nat. Mater. 2006, 5, 457.
[37] F. H. Gojny, M. H. G. Wichmann, B. Fiedler, I. A. Kinloch, W. Bau-
hofer, A. H. Windle, K. Schulte, Polymer 2006, 47, 2036.
www.advmat.de 2841