Acta Biomaterialia 18 (2015) 206212

Contents lists available at ScienceDirect

Acta Biomaterialia
journal homepage: www.elsevier.com/locate/actabiomat

Statistical shear lag model Unraveling the size effect in hierarchical

composites
Xiaoding Wei, Tobin Filleter 1, Horacio D. Espinosa
Department of Mechanical Engineering, Northwestern University, 2145 Sheridan Road, Evanston, IL 60208-3111, USA

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

Article history: Numerous experimental and computational studies have established that the hierarchical structures
Received 29 August 2014 encountered in natural materials, such as the brick-and-mortar structure observed in sea shells, are
Received in revised form 8 December 2014 essential for achieving defect tolerance. Due to this hierarchy, the mechanical properties of natural mate-
Accepted 28 January 2015
rials have a different size dependence compared to that of typical engineered materials. This study aimed
Available online 12 February 2015
to explore size effects on the strength of bio-inspired staggered hierarchical composites and to dene the
inuence of the geometry of constituents in their outstanding defect tolerance capability. A statistical
Keywords:
shear lag model is derived by extending the classical shear lag model to account for the statistics of
Size effect
Load transfer
the constituents strength. A general solution emerges from rigorous mathematical derivations, unifying
Shear lag the various empirical formulations for the fundamental link length used in previous statistical models.
Defect tolerance The model shows that the staggered arrangement of constituents grants composites a unique size effect
Hierarchical composites on mechanical strength in contrast to homogenous continuous materials. The model is applied to hierar-
chical yarns consisting of double-walled carbon nanotube bundles to assess its predictive capabilities for
novel synthetic materials. Interestingly, the model predicts that yarn gauge length does not signicantly
inuence the yarn strength, in close agreement with experimental observations.
2015 Acta Materialia Inc. Published by Elsevier Ltd. All rights reserved.

1. Introduction aws are randomly distributed along the strong and stiff laments,
and the number of aw scales with the dimensions of laments.
Hierarchical composite materials found in nature, such as nacre Thus, the stochastic nature of aw distribution needs to be taken
shells, collagen brils, and spider silk, are well known for their bal- into account in these continuum models. Following the pioneering
anced strength and toughness [16]. Through staggered arrange- work by Rosen and Zweben [16,17], various micromechanical
ments of constituents (i.e., strong and stiff laments embedded models based on the weakest-link formulations have been pro-
in softer interfaces), natural composites can not only preserve posed to predict strength distributions for unidirectional ber-re-
the stiffness and strength of the laments during material scale- inforced composites [1824]. All of them assumed that the
up, but also achieve remarkable toughness. Microstructure charac- failure probability for an individual lament of length L under
terizations of these natural composites reveal a critical length scale stress a follows Weibull statistics is given by
for each material that correlates to the two main failure mechan-   m 
L r
isms lament fracture and interface sliding to optimize macro- Pf r; L 1  exp  1
scopic strength and toughness simultaneously [710]. Different L 0 r0
continuum models have been advanced to relate critical length where r0 and m are the scale and shape parameters, respectively,
scales with mechanical properties of constituents, including key and L0 is the reference lament length. Curtin proposed the charac-
contributions by Gao et al. [11], Chen et al. [12], Rim et al. [13], teristic strength rc and length Lc [23], dened by
Wei et al. [14], and Barthelat et al. [15]. Yet, these analyses were !m=m1
 1=m1
conducted on models that had assumed only a specied aw with rm0 sL0 r0 rL1=m
0
a predened size and location. In practice, the location and size of rc and Lc 2
r s

Corresponding author. where r is the lament radius. Physically, Lc is twice the length
E-mail address: [email protected] (H.D. Espinosa).
around a lament break that is required for the stress in the la-
1
Current address: Department of Mechanical and Industrial Engineering, Univer- ment to recover linearly to the characteristic stress rc with the
sity of Toronto, 5 Kings College Rd., Toronto, ON M5S 3G8, Canada. assumption of a constant sliding resistance s (i.e. Lc rrc =s).

http://dx.doi.org/10.1016/j.actbio.2015.01.040
1742-7061/ 2015 Acta Materialia Inc. Published by Elsevier Ltd. All rights reserved.
 X. Wei et al. / Acta Biomaterialia 18 (2015) 206212 207

During studies of size effects in composite strength using statistical

models [20,21,2426], a fundamental link length ^L arose it is the
minimum link length below which the weakest-link scaling is not
applicable to predict the strength distribution of a realistic compos-
ite with a total length LT ii^L. Phoenix et al. suggest ^L aLc with
p
a  0:4 [21], and Landis et al. suggest ^L b 3pE=Gr with b  6
and E and G are the elastic modulus and shear modulus of the la-
ment and interface, respectively [25]. However, as noted in both
reports, the coefcients a and b may vary with the constituent
mechanical properties and arrangements, so they have to be deter-
mined through intensive Monte Carlo simulations.
In this study, by introducing the statistical shear lag model, we Fig. 1. Schematic of the 2D unit cell used in the statistical shear lag model
developed a mathematical framework that yielded a more general simplied from the brick-and-mortar structure of a hierarchical staggered
formulation. This formulation solves the ambiguity in the funda- composite.
mental link length scale and unies the previous models. Moreover,
the statistical shear lag model was implemented into a multiscale
statistical model to explain how staggered hierarchical structures the material dimension and loading conditions affect the structural
achieve superior aw tolerance. strength. Substituting Eq. (3) into Eq. (4), the probability of the rup-
ture of any two laments in the unit cell is given by
n RL
m o
2. Statistical shear lag model Pf rapp 1  exp  L20 0 rr1 x 0
dx
8 R
m 9
In our previous study, the classical shear lag model was used to < 2 L sinhkL=2
1 cosh 2k Lx sinhkx dx
= 5
0 2 rapp m
describe the optimal length scale in staggered composites that 1  exp  L0 r0
: ;
simultaneously optimize their strength and toughness [14]. As
shown in Fig. 1, a simplied 2D unit cell was constructed to study in which L0 is the reference length for the lament in the simplied
the load transfer efciency between two adjacent laments (with a 2D unit cell. By dening the effective length
thickness denoted as b) bonded by a thin soft interface layer (with Z     m
L
a thickness denoted as h) over the length L. A tensile normal stress 1 k kx
Leff L cosh L  x sinh dx; 6
rapp is applied at the right end of the top lament, and the left end 0 sinhkL=2 2 2
of the bottom lament is xed in the axial direction but is free to we can rewrite Eq. (5) as
deform in the lateral direction. The left end of the top lament    
and the right end of the bottom lament are traction-free and free 2Leff rapp m
Pf rapp 1  exp  : 7
to displace. Assuming the laments and interface deform elastical- L0 r0
ly, the solutions for axial tensile stress along the top and bottom
Eq. (7) suggests that the probability of the unit cell failing by
laments are expressed as
lament rupture follows a Weibull distribution, which possesses
kx

r1 x sinhrappkL sinh cosh kLx the same shape parameter as the Weibull distribution of the mono-
2 2
2

3 lithic lament rupture strengths. The new Weibull distribution for
kx
r2 x r1 L  x sinhrappkL cosh 2
sinh kLx
2
the rupture strength of the unit cell takes the form
2    
q rapp m
Pf rapp 1  exp  8
respectively. In Eq. (3), k 2G
Ehb
in which E is the elastic modulus of r0;uc
the lament and G is the shear modulus of the interface.
1=m
L0
Filament rupture and interface sliding were assumed to be the where r0;uc r0 2Leff
is the scale factor for the unit cell rupture
only two failure mechanisms that determine the composite strength.
strength. The classical shear lag model is based on several key The statistical shear lag model introduced above facilitates the
assumptions. First, the laments are treated as one-dimensional investigation of the statistics of unit cell rupture strengths as a
springs so that the transversal stress and variations in axial stress function of overlap length. For a lament material whose strength
are neglected. Second, the interfaces carry no tensile but only shear has a Weibull distribution, the unit cell rupture strengths also fol-
load controlled solely by the relative displacements of laments low Weibull statistics with a scale factor that is a function of the
adhered to the interface regions. Microscopic characterizations sug- overlap length, as shown in Eqs. (6) and (8). The variance of the
gest that the tensile strengths of reinforcements in natural compos- Weibull distribution for unit cell rupture strength is
ites spread widely due to variations in dominant defects [27]. In this r20;uc C1 2=m  C1 1=m2 . Alternatively, the unit cell can
study, we modied the classical shear lag model to account for the also fail by lamentlament sliding. For a linear elastic interface,
statistics in the tensile strength of reinforcements while keeping the unit cell fails in sliding when rapp 2sf =bk tanhkL=2, in
the key assumptions for the classical shear lag model. which sf is the interface shear strength [14]. Note that when scal-
Assuming the tensile strength of individual laments follows ing unit cells up to a macroscopic composite, the effective stress
the Weibull distribution function given by Eq. (1), the failure prob- applied on the unit cell is reff rapp b=2b h. When the interface
ability for a lament subject to a nonhomogeneous uniaxial stress
is thin (i.e., hhhb), reff equals approximately rapp =2.
state takes the integral form [28,29]:
To demonstrate the two competing failure mechanisms within
 Z   
rx m dLx the unit cell, a case study was performed on yarns made of dou-
Pf 1  exp  ; 4 ble-walled carbon nanotube (DWCNT) bundles (see Fig. 2a) [30].
L r0 L0
Each unit cell contains a pair of parallel DWCNT bundles. Statistics
where rx is the distribution function of the effective uniaxial of bundle strength have been obtained through previous in situ
stress over the sample length L. Clearly, Eq. (4) suggests that both SEM tensile experiments on 21 individual bundles with an average
208 X. Wei et al. / Acta Biomaterialia 18 (2015) 206212

Fig. 2. (a) Schematic depicting the hierarchical structure of a macroscopic DWCNT yarn. Each unit cell in the yarn consists of two parallel DWCNT bundles. (b) Weibull
analysis of the individual bundle rupture strength. (c) Simulated t using the shear lag model of experimental shear strengths measured for two parallel DWCNT bundles. (d)
Plot of the effective unit cell strength as a function of the unit cell length (overlap length). The unit cell strength determined by interface sliding is plotted in blue and by
bundle rupture is plotted in black. Each black square represents the scale factor of the Weibull distribution function of the unit cell rupture strength at a particular overlap
length. The error is the square root of the variance for each Weibull distribution function.

diameter of 25 nm and length of 5 lm [3033]. After ordering 21 shear lag model and the statistical shear lag model, the effects of
strength values ascendingly and dening P i 100i  0:5=21 as both failure modes on the unit cell strength are shown in Fig. 2d.
the sample percentile (i 1;    ; 21), the set of strength values The nal unit cell strength is the smaller of the two strength values
was tted with the Weibull distribution function given by Eq. (1) predicted by the statistical shear lag model and the classical shear
to yield a scale factor of r0 2:8 GPa and a shape factor m 2:2 lag model. The critical overlap length (or unit cell size) ~
L where the
(Fig. 2b). Note that the stress on a bundle in this study is dened classical and statistical shear lag model solutions intersect gives
as the load divided by the cross-sectional area of all carbon nan- the optimized unit cell strength and can be determined by
otube shells within the bundle. The mechanical properties of the numerically solving the following equation
interface were investigated through in situ SEM shear tests on
pairs of parallel bundles [32]. A pair of bundles aligned parallel !1=m !
L0 2sf k~L
to one another with various overlap lengths was pulled on the r0 tanh : 9
opposite ends of the bundles with the force to break the junction 2Leff ~L bk 2
recorded. As shown in Fig. 2c, tensile stresses required to slide
19 pairs of bundles were plotted as a function of overlap lengths. For unit cells smaller than ~
L, the scale factor r0;uc fails to describe
Previous coarse-grained simulations suggested that the interface the unit cell strength due to the interface sliding, and thus is invalid
between parallel bundles has a thickness h  2:5 nm [32]. The for weakest-link model for size-scaling the composite strength. In
equivalent lament thickness b is approximated as half of the the other words, ~ L has the same physical meaning as the aforemen-
bundle radius r. Fitting the results with the classical shear lag mod- tioned fundamental link length ^L. Moreover, we can show that Eq.
el solution rapp 2sf =bk tanhkL=2 yield an interface shear (9) is a more general solution that can be simplied to the formula-
strength sf 350 MPa and an effective interface shear modulus tions proposed previously [21,25] under specic conditions.
G 10 MPa. The properties of DWCNT bundles and interfaces are When m > 1, the effective unit cell length Leff given by Eq. (6) is
summarized in Table 1. Combining the results from the classical bounded by two limits

Table 1
Mechanical properties of DWCNT bundles and bundlebundle interface.

Bundle Equivalent half lament thickness, b Elastic modulus, E Strength scale factor, r0 Strength shape factor, Reference bundle length, L0
(nm) (GPa) (GPa) m (lm)

6.2 60 2.8 2.2 5

Interface Interface thickness, h (nm) Effective shear modulus, G (MPa) Shear strength, sf (MPa)
2.5 10 350
 X. Wei et al. / Acta Biomaterialia 18 (2015) 206212 209

Fig. 3. Schematic of the method used to scale up the statistical shear lag model. The unit cell for the shear lag model (top left) is simplied as an element (top right) that has
two failure modes (lamentlament sliding or lament rupture). A macroscopic hierarchical composite consisting of M  N unit cells (bottom left) is simplied into the
statistical model of M  N elements (bottom right). The cross-section is represented in the axial direction as a chain of N Daniels bundles, and in the lateral direction as M
parallel elements.

L L is modeled as a Daniels bundle of M parallel unit cells aligned in

> Leff > m : 10
m1 2 the lateral direction that are evenly clamped at both ends [34].
Along the axial direction, the composite is simplied as a chain
Leff approaches the upper limit for a small k (meaning the interface
of N Daniels bundles connected in series such that the composite
is very soft compared to the lament) and approaches the lower
has a total length Lcomp N  L. Daniels theory suggests that the
limit for a large k. Thus, for a small k, Eq. (9) can be approximated as
strength of a bundle consisting of M unit cells whose strength dis-
 1=m
L0 m 1 sf tribution is a Weibull probability function exhibits a Gaussian dis-
r0  ~L 11 tribution expressed as
2~L b
  
 m=m1 r l
r0 bL1=m m11=m Greff 12 1 erf peff2
Solving Eq. (11) yields ~
L sf
0
2
. Substituting 2d

the equivalent lament thickness b / r=2 for cylindrical laments in which;

 1   13
into this solution suggests that ~ L derived here has essentially the l m1 m exp  m1 r0;eff ;
same functional form as the one proposed by Phoenix et al. with  2    
d2 M1 m1 m exp  m1 1  exp  m1 r0;eff 2 ; and
only a slight change in the coefcient [21]. It is clear that as m
decreases, a larger critical link length ~L is required to apply weak- r0;eff r0;un 2bh
b

est-link scaling. This is consistent with the observation in the Mon-

The strength of a composite consisting of N serially connected
te Carlo simulations by Landis et al. [25]. Similarly, for a large k, Eq.
Daniels bundles is governed by the weakest bundle and thus is dis-
(9) can be approximated as
tributed as
 m 1=m
L0 2 2sf Hreff 1  1  Greff N : 14
r0 ~
 12
2L bk
q
m As shown by Phoenix and Raj [19], the Weibull approximation
Solving Eq. (12) yields another typical solution ~L rs0 2Gb L0
. to Hreff takes the form
Eh 2 f

m0 
This solution takes similar functional form as the one proposed by r
Landis et al., yet includes more constituent mechanical properties Hreff  1  exp  0 ; 15a
r
and geometries [25]. Therefore, through rigorous mathematical
derivation, the statistical shear lag model developed herein pro- with the composite Weibull scale factor r0 given by
vides a more general solution that solves the ambiguity in the fun- " #
q
loglogN log4p
damental link length scale obtained in previous statistical models r0 l d p  2 logN ; 15b
for ber-reinforced composites, bridging the gap between earlier 8 logN
theories.
and the composite shape factor m0 given by
p " #
3. Size effects in hierarchical composites 0 l 2 logN loglogN log4p
m 2 p  logN : 15c
d 2 logN
To extend the statistical shear lag model for evaluating the
strengths of hierarchical composites, the following strategy was
employed where the mechanical behavior of the unit cell was used 4. Results
as an input to investigate how the size of a macroscopic hierarchi-
cal composite is related to its strength. First, the unit cell (with Clearly, there is a size effect with respect to the strength of com-
length L P ~L) was simplied as an element consisting of two posites with different hierarchies as a function of the lament
sub-elements connected in series; one sub-element represents la- material (Eq. (1)), the unit cell (Eqs. 68), and the composite size
ment rupture, and the other represents lamentlament sliding (Eqs. (13), (14), (15a), (15b), (15c)). To illustrate this size effect, a
(see Fig. 3). Therefore, a hierarchical composite consisting of case study was performed on a DWCNT yarn consisting of unit cells
M  N unit cells can be simplied into the statistical model with an overlap length of 2.5 lm. The yarn contained 103 elements
consisting of M  N elements. The cross-section of the composite in the lateral direction and 106 Daniels bundles in the axial
210 X. Wei et al. / Acta Biomaterialia 18 (2015) 206212

Fig. 4. Probability density functions for different hierarchies in the DWCNT yarn. (a) The probability density functions of the DWCNT bundle rupture strength and unit cell
strength obtained from Eqs. (1) and (8); (b) the probability density functions of the Daniels bundle strength (consisting of 103 unit cells) and the macroscopic yarn strength
(consisting of 106 Daniels bundles) obtained from Eqs. (13) and (14).

direction. Fig. 4 shows the probability density distributions for the

strengths of 2.5 lm long and 30 nm diameter DWCNT bundle, a
unit cell, Daniels bundle of 103 elements (with a diameter of
approximately 30 lm), and a chain of 106 Daniels bundles (i.e. a
2.5 m long yarn). The Weibull distributions for DWCNT bundles
and unit cells share the same shape factor, while the scale factor
for the unit cell strength is lower than that for the DWCNT bundle
strength. The strength of the yarns cross section shows a Gaussian
distribution. In contrast, the strength distribution of a chain of 106
Daniels bundles exhibits a Weibull distribution function prole.
Fitting the data yields a scale factor of approximately 0.9 GPa and
a very high shape factor of 195, in excellent agreement with the
Weibull approximation (0.9 GPa and 203) given by Eqs. (15b) and
(15c). This much higher shape factor as compared with that of
2.2 for individual DWCNT bundles indicates an increase in reliabil-
Fig. 5. The predicted strength of the hierarchical DWCNT yarns as a function of yarn
ity as yarn length increases, which is consistent with previous
length demonstrates the suppressed length scale sensitivity. Experimentally
measured DWCNT yarn strength at various gauge lengths are included to compare reports [24].
with the model prediction (the error represents the standard deviation of yarn More interestingly, the statistical model explains why a
strength). hierarchical structure is so important in material scaling up from

Fig. 6. Schematic representation of multi-level hierarchical composites exhibiting a transition of strength statistics from Weibull to Gaussian to Weibull, at each hierarchical
level, during material scaling-up.
 X. Wei et al. / Acta Biomaterialia 18 (2015) 206212 211

nanoscale constituents to macroscopic composites. Fig. 5 shows that multiscale statistical model shows that the staggered composites
the mean strength value for the DWCNT yarn (a discontinuous hier- have suppressed size-dependent mechanical properties. This multi-
archical composite) depends barely on the yarn length when the scale statistical model reveals the transition of the statistics of
yarn length increases by 104 times (from 2.5 lm to 2.5  104 lm), strength during material scaling-up at each level in a composite
the mean yarn strength drops by only 10% (from 1 to 0.9 GPa). The with nested hierarchical structures. By factoring discontinuities into
model prediction agrees very well with the results of tensile experi- the composites in a controlled manner, staggered composites can
ments on DWCNT yarns [35]. The experimentally measured mean achieve an incredible capability for defect tolerance due to their
strengths at various yarn gauge lengths agree with the model pre- hierarchical structures. This nding provides valuable insights for
dictions and do not show appreciable length dependence. Insensi- the design of novel bio-inspired high-performance composite mate-
tivity of strength with respect to gauge length has also been rials. For example, with reasonable modications to account for the
observed in other composites containing natural or synthetic bers, difference in geometries of building blocks, this statistical model
such as sisal bers and carbon bers [26,36,37], as they share the can be applied to other nanocomposites, such as laminated gra-
same deformation mechanisms. By obtaining the mechanical prop- phene oxide-polymer composites, to guide the microstructure
erties of the constituents through micromechanical experiments, design that preserves the excellent mechanical properties of the
this model could potentially be applied to those composites to gain nanoscale building blocks [3941].
further insights into the microstructureproperty relationships.
Thus, understanding why the hierarchical structures are common Acknowledgements
in most of the natural composites is straightforward through this
approach, which reveals the ability of discontinuous staggered com- The authors acknowledge support from ARO through MURI
posites to re-distribute load over characteristic length scales that award W911NF-09-1-0541 and the NSF through DMREF award
are functions of constituent geometry and mechanical properties. CMMI-1235480. The authors also thank Matthew Ford for assis-
Hence, macroscopic natural composites can preserve remarkable tance in equation derivation and discussions.
mechanical properties when scaling up from the laments usually
in the nm or lm length scales.
Appendix A. Figures with essential colour discrimination

5. Discussion
Certain gures in this article, particularly Figs. 16, are difcult
to interpret in black and white. The full colour images can be found
Finally, it is worth to note that the statistical models previously
in the on-line version, at http://dx.doi.org/10.1016/j.actbio.2015.
introduced can be applied to composites with nested hierarchical
01.040.
structures (as shown in Fig. 6). With the strength of the building
block at level 0 following a Weibull distribution, based on the
References
above result, at level 1 the statistics of strength starts with a Wei-
bull distribution for the unit cell, then transitions to a Gaussian dis- [1] Currey J. Mechanical properties of mother of pearl in tension. Proc R Soc Lond B
tribution when the material scales in the lateral direction, and Biol Sci 1977;196:44363.
nally returns to a new Weibull distribution when the material [2] Barthelat F, Tang H, Zavattieri P, Li C-M, Espinosa H. On the mechanics of
mother-of-pearl: a key feature in the material hierarchical structure. J Mech
scales in the axial direction (to form a bril). It is obvious that Phys Solids 2007;55:30637.
the bril at the end of the hierarchical level 1 serves as reinforce- [3] Fratzl P, Misof K, Zizak I, Rapp G, Amenitsch H, Bernstorff S. Fibrillar structure
ment for the unit cell at level 2. According to the statistical shear- and mechanical properties of collagen. J Struct Biol 1998;122:11922.
[4] Keten S, Xu Z, Ihle B, Buehler MJ. Nanoconnement controls stiffness, strength
lag model, the strength of the unit cell at level 2 should follow a and mechanical toughness of [beta]-sheet crystals in silk. Nat Mater
Weibull distribution as well. Replicating this procedure suggests 2010;9:35967.
that this WeibullGaussianWeibull distribution transition repeats [5] Espinosa HD, Rim JE, Barthelat F, Buehler MJ. Merger of structure and material
in nacre and bone perspectives on de novo biomimetic materials. Prog Mater
at each hierarchical level according to Eqs. 6, 7, 8, 13, 14, 15a, 15b Sci 2009;54:1059100.
and 15c. Therefore, the statistical model developed in the current [6] Espinosa HD, Juster AL, Latourte FJ, Loh OY, Gregoire D, Zavattieri PD. Tablet-
study is a self-consistent multi-scale model that can predict the level origin of toughening in abalone shells and translation to synthetic
composite materials. Nat Commun 2011;2:173.
mechanical strength of bio-inspired hierarchical composites and
[7] Meyers MA, Lin AY-M, Chen P-Y, Muyco J. Mechanical strength of abalone
reveal the unique size-effect on the mechanical strength that ben- nacre: role of the soft organic layer. J Mech Behav Biomed Mater
ets from the staggered hierarchical structure. We note that the 2008;1:7685.
statistical shear lag model introduced here assumed an ideally par- [8] Landis W, Song M, Leith A, McEwen L, McEwen B. Mineral and organic matrix
interaction in normally calcifying tendon visualized in three dimensions by
allel alignment of laments. In the future, effects of lament bend- high-voltage electron microscopic tomography and graphic image
ing and torsion will be taken into account so that the model can be reconstruction. J Struct Biol 1993;110:3954.
applied to composites that have more complicated microstructures [9] Penel S, Morrison RG, Dobson PD, Mortishire-Smith RJ, Doig AJ. Length
preferences and periodicity in b-strands. Antiparallel edge b-sheets are more
(e.g., CNT ropes containing hierarchical helical microstructures) likely to nish in non-hydrogen bonded rings. Protein Eng 2003;16:95761.
[38]. [10] Shao Y, Zhao H-P, Feng X-Q, Gao H. Discontinuous crack-bridging model for
fracture toughness analysis of nacre. J Mech Phys Solids 2012;60:140019.
[11] Gao H, Ji B, Jger IL, Arzt E, Fratzl P. Materials become insensitive to aws at
6. Concluding remarks nanoscale: lessons from nature. Proc Natl Acad Sci 2003;100:5597600.
[12] Chen B, Wu P, Gao H. A characteristic length for stress transfer in the
In the current study, the statistical shear lag model was extended nanostructure of biological composites. Compos Sci Technol 2009;69:11604.
[13] Rim JE, Zavattieri P, Juster A, Espinosa HD. Dimensional analysis and
to investigate the size effect on the hierarchical composites parametric studies for designing articial nacre. J Mech Behav Biomed Mater
mechanical properties at all levels including the laments, unit cell, 2011;4:190211.
and the macroscopic composite. The model suggests that the statis- [14] Wei X, Naraghi M, Espinosa HD. Optimal length scales emerging from shear
load transfer in natural materials: application to carbon-based nanocomposite
tics of the lament strength plays an important role in the optimiza-
design. ACS Nano 2012;6:233344.
tion of the geometries of the unit cell. A more general solution [15] Barthelat F, Dastjerdi AK, Rabiei R. An improved failure criterion for biological
emerged for the critical link length scale beyond which the Weibull and engineered staggered composites. J R Soc Interface 2013;10:20120849.
analysis can be applied to predict size scaling of composite strength. [16] Rosen BW. Tensile failure of brous composites. AIAA J 1964;2:198591.
[17] Zweben C. Tensile failure of ber composites. AIAA J 1968;6:232531.
This solution unies the formulations proposed in previous statisti- [18] Curtin WA. Theory of mechanical properties of ceramic-matrix composites. J
cal models that were determined empirically. Furthermore, the Am Ceram Soc 1991;74:283745.
212 X. Wei et al. / Acta Biomaterialia 18 (2015) 206212

[19] Phoenix S, Raj R. Overview no. 100 scalings in fracture probabilities for a brittle [31] Filleter T, Bernal R, Li S, Espinosa H. Ultrahigh strength and stiffness in cross-
matrix ber composite. Acta Metall Mater 1992;40:281328. linked hierarchical carbon nanotube bundles. Adv Mater 2011;23:285560.
[20] Curtin W, Takeda N. Tensile strength of ber-reinforced composites: II. [32] Naraghi M, Bratzel GH, Filleter T, An Z, Wei X, Nguyen ST, et al. Atomistic
Application to polymer matrix composites. J Compos Mater 1998;32:206081. investigation of load transfer between DWNT bundles Crosslinked by PMMA
[21] Phoenix S, Ibnabdeljalil M, Hui C-Y. Size effects in the distribution for strength oligomers. Adv Funct Mater 2013;23:188392.
of brittle matrix brous composites. Int J Solids Struct 1997;34:54568. [33] Papkov D, Beese AM, Goponenko A, Zou Y, Naraghi M, Espinosa HD, et al.
[22] Batdorf S, Ghaffarian R. Size effect and strength variability of unidirectional Extraordinary improvement of the graphitic structure of continuous carbon
composites. Int J Fract 1984;26:11323. nanobers templated with double wall carbon nanotubes. ACS Nano
[23] Curtin W. Ultimate strengths of bre-reinforced ceramics and metals. 2012;7:12642.
Composites 1993;24:98102. [34] Daniels H. The statistical theory of the strength of bundles of threads. I. Proc R
[24] Ibnabdeljalil M, Curtin W. Strength and reliability of ber-reinforced Soc Lond A Math Phys Sci 1945;183:40535.
composites: localized load-sharing and associated size effects. Int J Solids [35] Beese AM, Wei X, Sarkar S, Ramachandramoorthy R, Roenbeck MR, Moravsky
Struct 1997;34:264968. A, et al. Key factors limiting carbon nanotube yarn strength: exploring
[25] Landis CM, Beyerlein IJ, McMeeking RM. Micromechanical simulation of the processingstructureproperty relationships. ACS Nano 2014;8:1145466.
failure of ber reinforced composites. J Mech Phys Solids 2000;48:62148. [36] Silva FdA, Chawla N. Tensile behavior of high performance natural (sisal)
[26] Okabe T, Takeda N. Size effect on tensile strength of unidirectional CFRP bers. Compos Sci Technol 2008;68:343843.
compositesexperiment and simulation. Compos Sci Technol 2002;62: [37] Pickering K, Murray T. Weak link scaling analysis of high-strength carbon bre.
205364. Compos A Appl Sci Manuf 1999;30:101721.
[27] Menig R, Meyers MH, Meyers MA, Vecchio KS. Quasi-static and dynamic [38] Zhao Z-L, Zhao H-P, Wang J-S, Zhang Z, Feng X-Q. Mechanical properties of
mechanical response of Haliotis rufescens (abalone) shells. Acta Mater carbon nanotube ropes with hierarchical helical structures. J Mech Phys Solids
2000;48:238398. 2014;71:6483.
[28] Bazant ZP, Planas J. Fracture and size effect in concrete and other quasibrittle [39] Stankovich S, Dikin DA, Dommett GH, Kohlhaas KM, Zimney EJ, Stach EA, et al.
materials. CRC press; 1997. Graphene-based composite materials. Nature 2006;442:2826.
[29] Bazant Z. Size effect on structural strength: a review. Arch Appl Mech [40] Beese AM, An Z, Sarkar S, Nathamgari SSP, Espinosa HD, Nguyen ST. Defect-
1999;69:70325. tolerant nanocomposites through bio-inspired stiffness modulation. Adv Funct
[30] Naraghi M, Filleter T, Moravsky A, Locascio M, Loutfy RO, Espinosa HD. A Mater 2014;24:288391.
multiscale study of high performance double-walled nanotube  polymer [41] Suk JW, Piner RD, An J, Ruoff RS. Mechanical properties of monolayer graphene
bers. ACS Nano 2010;4:646376. oxide. ACS Nano 2010;4:655764.