Blending stiffness and strength disorder can stabilize fracture

Ehud D. Karpas
Weizmann Institute of Science, Rehovot, 7610001, Israel

Ferenc Kun∗
Department of Theoretical Physics, University of Debrecen, P.O.Box:5, H-4010 Debrecen, Hungary
(Dated: April 7, 2016)
Quasi-brittle behavior where macroscopic failure is preceded by stable damaging and intensive
arXiv:1604.01430v1 [cond-mat.mtrl-sci] 5 Apr 2016

cracking activity is a desired feature of materials because it makes fracture predictable. Based on a
fiber bundle model with global load sharing we show that blending strength and stiffness disorder of
material elements leads to the stabilization of fracture, i.e. samples which are brittle when one source
of disorder is present, become quasi-brittle as a consequence of blending. We derive a condition of
quasi-brittle behavior in terms of the joint distribution of the two sources of disorder. Breaking
bursts have a power law size distribution of exponent 5/2 without any crossover to a lower exponent
when the amount of disorder is gradually decreased. The results have practical relevance for the
design of materials to increase the safety of constructions.

PACS numbers: 64.60.av,46.50.+a,81.40.Np

I. INTRODUCTION the local stiffness can improve the fracture toughness of

materials [15, 16].
Disorder is an inherent property of practically all ma- Beyond artificially made (tailored) materials, struc-
terials, both natural and man made. The heterogeneity tural heterogeneity of local stiffness is an important fea-
occurring on different length scales plays a crucial role ture of biological materials, as well [17–20]. One of the
in the mechanical response and fracture behavior of ma- most known example of such biological composite ma-
terials. On the one hand, the presence of flaws, voids, terials is nacre which exhibits extraordinary mechanical
and grain boundaries reduces the fracture strength, on properties compared to its constituent materials. These
the other hand, however, they improve the damage toler- features can be partly attributed to the interplay of the
ance of materials which has a high practical importance local variation of stiffness and strength [17, 18].
for construction components [1–3]. Low disorder leads to In the present paper we consider this problem in the
brittle behavior where fracture occurs at a critical load in framework of fiber bundle models (FBM) by introduc-
a catastrophic manner without any precursors. From a ing two sources of disorder, i.e. both the stiffness and
practical point of view quasi-brittle behavior is desired, strength of fibers are random variables. Assuming global
which is typical for materials with a higher amount of load sharing after failure events we obtain a generic ana-
disorder. The fracture process of quasi-brittle materials lytical description of the mechanical response of the sys-
is composed of a large number of intermittent steps of tem on the macro-level, and investigate the microscopic
cracking giving rise to the emergence of crackling noise process of failure by computer simulations. We show that
[4, 5]. Analyzing the statistics and dynamics of crackling blending stiffness and strength disorder results in stabi-
noise, methods can be worked out to forecast the immi- lization of fracture even if the system with a single source
nent global failure [6]. Controlling the amount of disorder of disorder had a perfectly brittle response.
to enhance the quasi-brittle behavior of materials has a
high practical relevance.
The theoretical investigation of the fracture of hetero- II. FIBER BUNDLE WITH TWO RANDOM
geneous materials is mainly based on discrete stochastic FIELDS
models such as Fiber Bundle (FBM) [7–10], fuse model
[11] and Discrete Element (DEM) approaches [12–14] In the model we consider a parallel set of N fibers
with Monte Carlo and molecular dynamics simulation which is loaded along the fibers’ direction. Fibers have a
techniques. A common basic assumption of such mod- perfectly brittle response, i.e. they exhibit a linearly elas-
eling approaches is that the heterogeneity of materials tic behavior with a Young modulus E and fail when the
can be fully captured by introducing disorder for the local load σ on them exceeds a threshold value σth . In order
fracture strength of cohesive elements of the model. How- to capture the local variation of stiffness and strength
ever, recent experimental and theoretical investigations of materials, it is a crucial element of the model that
led to the surprising conclusion that the heterogeneity of each fiber is characterized by two random variables: the
Young modulus of fibers E takes values in the inter-
val E− ≤ E ≤ E+ according to the probability distri-
bution f (E), while the breaking threshold σth is sam-
∗ ferenc.kun@science.unideb.hu pled with the probability density g(σth ) over the interval
 2

σ− ≤ σth ≤ σ+ . In the present study the two random

fields are assumed to be independent, i.e. no correlation
is considered between strength and stiffness. Hence, in

+
a finite bundle of N fibers two independent random val-
i
ues σth and Ei are assigned to each fiber i = 1, . . . , N .
When fibers fail during the stress controlled loading of Intact fibers
the bundle the load of broken fibers has to be overtaken

th
by the remaining intact ones. For simplicity, we assume
infinite range of load sharing which can be ensured by
loading the bundle between two perfectly rigid platens.
It has the consequence that the strain ε of fibers is always
the same, however, due to the randomness of the Young

-
modulus E their local load σi is a fluctuating quantity
σi = Ei ε, where i = 1, . . . , N .
In the classical FBM [9, 10, 21] the strength of fibers
σth is the only random variable which represents the
heterogeneity of materials. Since the Young modulus is E- E E+
constant E = const., in the limit of global load sharing
of the model fibers keep the same load Eε, and hence,
break in the increasing order of their breaking threshold FIG. 1. Failure plane of fibers for uniformly distributed
i
σth , i = 1, . . . , N . It has the consequence that the macro- threshold values. Each point with parameter values (E, σth )
scopic constitutive equation σ0 (ε) can be expressed in inside the rectangle of side lengths E+ − E− and σ+ − σ−
represents a fiber of the bundle. The equation of the dashed
terms of the cummulative distribution of strength thresh-
Rσ straight line is σth = Eε so that fibers above the line full-
olds G(σth ) = σ−th g(x)dx in the form fill the condition σth > Eε. At a given strain ε during the
loading process those fibers are intact (highlighted with gray
σ0 (ε) = Eε [1 − G(Eε)] . (1) color) and keep the external load, which fall above this line
of slope ε. The integral in Eq. (3) has to be performed over
Here the term [1 − G(Eε)] provides the fraction of intact the domain of intact fibers.
fibers at the strain ε, and σ0 denotes the external load.
In the opposite limit of the model all fibers have the
same breaking threshold σth = const., however, their load sharing fibers break in the increasing order of the
Young modulus E is random with Ei , i = 1, . . . , N val- local failure strain εith (i = 1, . . . , N ).
ues. The breaking condition Ei ε > σth implies that in Since the stiffness and the failure strength are inde-
this case fibers break in the decreasing order of their pendent random variables, the load carried by the intact
Young moduli Ei . Recently, we have shown that in this fibers having Young modulus and strength in the interval
case the constitutive equation of the model can be ob- [E, E + dE] and [σth , σth + dσth ], respectively, reads as
tained as [22] Eεf (E)g(σth )dEdσth . Integrating the contributions of
all intact fibers we obtain the generic form of the consti-
Z σth /ε
tutive equation
σ0 (ε) = ε Ef (E)dE, (2)
E− Z σ+ Z σth /ε
σ0 (ε) = ε Ef (E)g(σth )dEdσth , (3)
where f (E) is the probability density of the Young modu- σ− E−
lus of fibers. Equation (2) expresses that at a given strain
ε those fibers are intact in the bundle whose stiffness E in terms of the probability density functions f and g of
falls below σth /ε. the Young modulus and strength of fibers, respectively.
Our present fiber bundle model is a combination of First the integral over E has to be performed, where the
the above two cases allowing for randomness both in the upper limit σth /ε of the integral captures the effect that
stiffness E and strength σth of fibers. In the presence at the macroscopic strain ε only those fibers can be intact
of two disorder fields E and σth the breaking sequence which have a Young modulus below σth /ε (similarly to
becomes more complex: Fibers break when the load on Eq. (2)). Then the integral over the strength σth of single
them Ei ε exceeds the local breaking threshold Ei ε > σth i
, fibers follows, where σth can take any value in the range
hence, at a strain ε those fibers are broken for which the σ− ≤ σth ≤ σ+ .
condition ε > σth i
/Ei holds. It can be seen that the It can be observed that taking the small strain limit
breaking sequence of fibers is controlled by the ratio of ε → 0 in the constitutive equation Eq. (3) we restore
their strength and stiffness which defines their critical linear behavior in the form σ0 (ε → 0) = ε hEi where
strain of breaking εith = σth
i
/Ei , and hence, the breaking hEi denotes the average Young modulus of fibers hEi =
R E+
condition can be formulated as ε > εith . It follows that in E− Ef (E)dE. In the large strain limit the macroscopic
our model of random stiffness and strength with global stress goes to zero σ0 (ε → ∞) → 0 since there are no
 3

intact fibers left. It is important to note that setting the For brevity, we define the notation
probability distribution of the Young modulus or break- B ≡ f (E)g(σth ) = 1/[(E+ − E− )(σ+ − σ− )], and the
ing threshold to a Dirac delta function f (E) = δ(E − E0 ) strains where the first and last fibers break are denoted
0
and g(σth ) = δ(σth − σth ), the constitutive equation Eq. by εmin ≡ σ− /E+ , εmax ≡ σ+ /E− , respectively. In
(3) of our model recovers the FBM equations Eqs. (1) and addition, we introduce ε1 ≡ σ− /E− , ε2 ≡ σ+ /E+
(2) with only one source of disorder for failure strength and point out that ε1 can be smaller or larger than ε2
[9, 10, 21] and for the Young modulus [22], respectively. depending on the parameters of the density functions
In the following we investigate the breaking process of Eq. (4).
our FBM both on the macro and micro scales. In order
to clarify the effect of blending strength and stiffness dis-
order we focus on systems which exhibit perfectly brittle
failure if only one source of disorder is present.
A. Macroscopic response

III. UNIFORMLY DISTRIBUTED STIFFNESS

The macroscopic constitutive curve σ0 (ε) of the bun-
AND STRENGTH
dle can be obtained by inserting the above probability
densities Eq. (4) into the generic form Eq. (3). Figure 1
The details of the macroscopic response of the system illustrates the failure plane (σth , E) where the breaking
and of the breaking process of fibers can easily be deter- thresholds εth of fibers can take values. When the strain ε
mined analytically when both the Young modulus E and is reached during the loading process those fibers remain
the strength σth of fibers are uniformly distributed with intact and keep the load which fall above the straight line
the probability densities of slope ε (it is highlighted with gray color in the figure).
1 1 Assuming the case ε1 < ε2 the integrals of Eq. (3) can
f (E) = , and g(σth ) = . (4) be carried out in a piecewise manner as
E+ − E− σ+ − σ−



 0.5(E+ + E− )ε, ε < εmin ;


3
1 σ−


 3 2 2 2 2

 [−2E + ε − + 3(E+ σ+ + E− σ− − E− σ+ )ε], εmin < ε < ε1 ;
6B ε



1

σ0 (ε) = [−2(E+3 3
− E− )ε2 + 3σ+ (E+2
− E−2
)ε], ε1 < ε < ε2 ; (5)

 6B
σ3

 1

3 2
[2E− ε − 3σ+ E− 2
ε + + ], ε2 < ε < εmax ;


6B ε





0, εmax < ε.


If ε2 < ε1 the macroscopic behavior is the same except emerges as the consequence of gradual breaking of fibers
for the interval ε1 < ε < ε2 (the third interval of Eq. (5)) indicating the quasi-brittle behavior of the bundle. Note
which is replaced by that the decreasing regime of the constitutive curves in
Fig. 2 can only be realized under strain controlled load-
1 3 3 1 2 ing conditions. Subjecting the system to an increasing
σ0 (ε) = [−2(σ+ − σ− ) − 3(σ+ − σ− )E− ε]. (6)
6B ε external load catastrophic collapse occurs when the peak
In this case we must also exchange ε1 and ε2 every- of σ0 (ε) is surpassed. The value σ0c of the peak stress
where in the limits of the intervals. In order to quantify defines the fracture strength of the bundle, while the
the amount of disorder in the system, without loss of gen- peak position εc provides the critical strain. It can be
erality, from here on end we fix the upper limits E+ = 1 observed in Fig. 2 that the extension of the non-linear
and σ+ = 1 and control the disorder by the width of the regime preceding macroscopic failure, and hence, the de-
distributions WE and Wσ such that WE ≡ E+ − E− and gree of brittleness, strongly depends on the amount of
Wσ ≡ σ+ − σ− . The macroscopic response σ0 (ε) of the disorder.
fiber bundle is presented in Fig. 2 for several values of
Wσ keeping the width WE = 0.5 fixed. It can be ob-
served that for strains ε < εmin , where no fiber breaking B. Fraction of broken fibers
occurs, the system exhibits a perfectly linear response
with an effective Young modulus equal to the average In order to quantify the degree of brittleness of the sys-
value hEi = (E+ + E− )/2 of E. Above εmin non-linearity tem we determined the fraction of fibers Pb which break
 4

0.75 0.5

WE=0.5 W =0.1
0.6 0.4

Pb
0.5 0.4 0.3
W =0.67
0

0.2
0.2
0.25 0
1 0.1
1
0.5 0.5
W =0.9 0
WE 0 0
0.0
0.0 0.5 1.0 1.5 2.0 Wσ
FIG. 3. (Color online) The fraction of fibers Pb which break
before the peak of the constitutive curve is reached dur-
FIG. 2. (Color online) Macroscopic response of a system
ing stress controlled loading of the bundle. The surface of
where both sources of disorder are uniformly distributed with
Pb (WE , Wσ ) was obtained by numerically evaluating the an-
the parameter WE = 0.5 for three different values of Wσ .
alytical expression Eq. (7). Pb has a finite value everywhere
Different symbols and colors are used to highlight the regimes
on the WE - Wσ plane except on the two axis.
corresponding to different terms of the integral expression Eq.
(5): bold line (red), circle (blue), square (green), and trian-
gle (magenta) stand for the contributions of the first, second,
third, and fourth terms of Eq. (5). perfectly linear up to the maximum.
For the case of WE = 0 the integral of Eq. (7) can
be carried out analytically, which yields for the breaking
before the collapse at σ0c under stress controlled loading. fraction
Perfectly brittle behavior is characterized by the value 
 0 Wσ < 0.5,
Pb = 0, since in this case the breaking of the first fiber
Pb (WE = 0, Wσ ) = 1 (8)
gives rise to catastrophic failure of the system. Starting  1− , 0.5 ≤ Wσ ≤ 1.
from the constitutive equation Eq. (5) we can find the 2Wσ
critical strain εc , and then Pb can be obtained from the
This result shows that in the absence of stiffness dis-
disorder distribution as order WE = 0 a transition occurs at Wσ = 1/2 between
Z σ+ Z σth /εc a perfectly brittle Pb = 0 and a quasi-brittle behavior
Pb = 1 − f (E)g(σth )dEdσth . (7) Pb > 0. The numerical results demonstrate that for any
σ− E−
finite amount of stiffness disorder WE > 0 the transi-
It has been shown analytically in the classical fiber tion disappears since always a finite fraction of fibers
bundle model, i.e. in the absence of stiffness disorder, breaks before failure Pb > 0. It follows that blending
that the system has a perfectly brittle response for nar- stiffness and strength disorder can stabilize the system in
row distributions Wσ ≤ 0.5 of fibers’ strength [23, 24]. the sense that no catastrophic collapse can occur without
Recently, we have demonstrated that in the opposite precursors.
limit when the uniformly distributed random stiffness is
the only source of disorder, the macroscopic response of
the bundle is perfectly brittle at any value of WE [22]. IV. CONDITION OF STABILITY
We evaluated the integral of Eq. (7) by numerical means
varying the amount of disorder over the entire range In the previous section it has been shown using the cu-
0 ≤ WE , Wσ ≤ 1. It can be seen in Fig. 3 that Pb obtains mulative quantity Pb that mixing stiffness and strength
a finite value Pb > 0 everywhere in the WE - Wσ plane disorder results in stability of the system in the sense
except for the WE axis where Wσ = 0 holds, and in the that immediate catastrophic failure at the time of first
range Wσ ≤ 1/2 on the Wσ axis where WE = 0 holds. breaking is avoided. In the following we analyze the tran-
The result has the astonishing consequence that when- sition from the perfectly brittle to quasi-brittle behavior
ever there is disorder present both in local strength and by focusing on the microscopic dynamics of the failure
stiffness of material elements the macroscopic response of process.
the system is quasi-brittle, i.e. a finite fraction of fibers We can formulate a criterion for the stability of the
breaks before the catastrophic failure of the bundle, so fracture process in terms of the disorder distributions
that the constitutive curve of the system σ0 (ε) is never based on the idea that the system is perfectly brittle if
 5
2.5
WE=0.98 W =0.99
the first fiber breaking induces a catastrophic avalanche. WE=0.72 W =0.83
At the breaking of the first fiber with the threshold value 2.0 WE=0.3 W =0.4
εmin
th = σ− /E+ the load on the bundle is hEi εmin th =
hEi σ− /E+ . After the breaking event the new Young 1.5

th)
′
modulus can be approximated as hEi ≈ hEi − E+ /N ,
′
which gives rise to a higher strain ε of the bundle

h(
  1.0
σ− hEi
ε′ = . (9)
E+ hEi − E+ /N 0.5
Consequently, the strain increment ∆ε = ε − ′
εmin
gen-th
erated by the breaking event under a fixed load can be 0.0
0 1 2 3
cast in the form
th
σ−
∆ε = . (10)
N hEi
FIG. 4. (Color online) Probability distribution h(εth ) of the
The average number of fiber breakings a(εmin
th ) induced strain thresholds εth of fibers calculated from Eq. (15) for
by the first failure reads as three parameter sets.
 
σ− σ−
a(εmin
th ) = N h(ε min
th )∆ε = h , (11)
E+ hEi
ε1 < ε2 holds
where h(εth ) denotes the probability distribution of  σ−
threshold strains εth . The avalanche induced by the 0
 εth < ,
first fiber breaking becomes catastrophic if a(εmin E+
th ) > 1,




which yields the stability condition of our system σ2
 
1 σ− σ−


2
E+ − 2− , ≤ εth < ,



  
 2B εth E+ E−
σ− σ−


h < 1. (12)

 1  2 2 σ− σ+
E+ hEi

h(εth ) = E+ − E− , ≤ εth < , (15)

 2B E− E+
Note that the condition is general and can be applied to
  2
σ+
 

 1 2 σ+ σ+
any disorder distribution. − E− , ≤ εth < ,



2B 2
εth E+ E−
i

If there is only stiffness disorder present (σth = σth ,


σ+


i = 1, . . . , N ) the distribution h(εth ) can be obtained 
0
 εth > .
from the stiffness distribution f (E) with a simple trans- E−
formation
Figure 4 illustrates the distribution h(εth ) for three com-
binations of WE and Wσ . It can be observed that for
 
σth σth
h(εth ) = f . (13) uniformly distributed stiffness and strength the distribu-
εth ε2th
tion of the strain thresholds h(εth ) starts continuously
Substituting e.g. the uniform distribution√ Eq. (4) we ob- from zero for any finite value of σ− without any finite
tain the condition E+ /(E+ − E− ) < 1/ 2, which can jump. This feature explains why the combination of two
never hold. It follows that if the stiffness disorder is the perfectly brittle systems leads to the emergence of quasi-
only source of heterogeneity of the system, for uniformly brittle behavior where stable cracking precedes macro-
distributed stiffness values the system always has a per- scopic failure. For the case σ− = 0 the distribution
fectly brittle behavior. In Ref. [22] the same result was starts with a finite constant, however, the small strain
obtained but in a different way focusing on the shape of value ε ≃ 0 still ensures stability. For those disorder dis-
the constitutive curve. tributions of stiffness and strength which cover the range
When both the stiffness and strength of fibers have from 0 to +∞ stability of the blend is again guaranteed
disorder the probability distribution of the strain thresh- by the generic form of the distribution Eq. (14).
olds h(εth ) can be calculated as the convolution of f (E)
and g(σth ) taking the ratio of the two random variables
εth = σth /E V. AVALANCHES OF FIBER FAILURES

Z E+
Under quasi-statically increasing external load σ0
h(εth ) = Ef (E)g(εth E)dE. (14)
E− when a fiber breaks its load gets redistributed over the
remaining intact fibers which may induce further fail-
We carried out the integration for the specific case when ure events. As a consequence of subsequent load redis-
both random variables are uniformly distributed and tribution a single breaking fiber may trigger an entire
 6

′
avalanche of breaking events. The randomness of local Since hEi < hEi and the external load is kept con-
physical properties and the interaction of fibers intro- stant, the strain of the bundle increases to the new
duced by the load sharing result in highly complex mi- value ε′ , which can be obtained as
croscopic dynamics of the failure process [9, 25]. In the
′
following we explore the statistics of breaking avalanches ε′ = hEi ε/ hEi . (18)
of fibers by computer simulations.
5. Those fibers which have threshold values below the
the updated strain εjth < ε′ have to be removed
A. Computer simulation technique and the algorithm is continued with step 4. During
bursts of breaking one has to take into account in
First we present the algorithm which allows us to sim- Eq. (17) that more than one fiber may also break
ulate the fracture process of large bundles. It has been in an iteration step.
assumed that the bundle is loaded between stiff platens 6. If no more fibers break due to load redistribution,
which ensures that the strain of fibers is the same ε. As the avalanche ended and the external strain can be
the external load σ0 is increased the fibers break in the increased again to the strain threshold of the next
increasing order of their failure strain εith , determined intact fiber in the sorted sequence of εth .
as εith = σth
i
/E i (i = 1, . . . , N ) which fall in the range
σ− /E+ ≤ εth ≤ σ+ /E− . Computer simulation of the The efficiency of the algorithm enabled us to simulate
failure process of a finite bundle of N fibers under stress bundles of N = 107 fibers averaging over 5000 samples
controlled loading proceeds in the following steps: at each parameter set with moderate CPU times.

1. Generate random values of the Young modulus Ei ,

i
and failure strength σth , i = 1, . . . , N of fibers B. Size distribution of bursts
according to the desired distributions f (E) and
g(σth ). Independence of the random fields has to Using the above algorithm we explored the bursting
be ensured. activity accompanying the stress controlled loading pro-
cess of quasi-brittle systems with two sources of disorder.
2. Determine the failure strains εith = σth i
/Ei , i =
The avalanches are characterized by their size ∆, which
1, . . . , N , and sort them into ascending order.
is defined as the number of fibers breaking in the cor-
related trail of the avalanche. For the classical FBM,
3. Increase the externally imposed strain ε up to the
where the fiber strength is the only source of disorder, it
smallest threshold ε = εmin
th , and remove the break- has been shown analytically [9, 10, 25] and by computer
ing fiber. At this instant there is σ0 = hEi ε load on
simulations [9, 26] that for equal load sharing the size
the system, where the initial value of the average
distribution of avalanches p(∆) has a power law behav-
Young modulus of the bundle reads as
ior
N
1 X p(∆) ∼ ∆−τ . (19)
hEi = Ei . (16)
N i=1
The value of the exponent τ = 5/2 is universal for a
broad class of failure threshold distributions [9, 10, 25].
4. After the breaking event the load of the broken Avalanches occur until the disorder is high enough in
fiber gets redistributed over the remaining intact the system Wσ > σ+ /2 [23, 24]. In the limiting case
ones keeping the external load σ0 constant. This Wσ → σ+ /2 the quasi-brittle region where avalanche
load redistribution may induce additional fiber fail- precursors occur, shrinks such that when Wσ ≤ σ+ /2
ures and eventually can even trigger an avalanche the response of the bundle becomes perfectly brittle
of breaking events. Note that due to the random and the system collapses without having any finite size
stiffness, fibers keep a different amount of load that avalanches. The constitutive behaviour of the system
is why long range interaction is easier to realize under stress controlled loading is shown in the inset of
through the control of strain. Triggered breakings Fig. 5 for a few values of Wσ , where the shrinking of the
can be determined in the following way: after the quasi-brittle regime can be observed. Approaching the
breaking event the overall Young modulus of the brittle limit the power law size distribution of avalanches
bundle has to be updated prevails, however, p(∆) exhibits a crossover to a lower
exponent τ = 3/2 in agreement with previous findings
1 X
hEi′ = Ei , (17) [9, 23, 24]. This behavior can be seen in Fig. 5 where
Nint avalanche size distributions p(∆) of our model are pre-
i∈I
sented for several values of Wσ at zero stiffness disorder
where I denotes the set of intact fibers which has WE = 0. Approaching the brittle limit the lower value
Nint elements. of the avalanche size exponent τ = 3/2 means that the
 7

10 0.5 exponent τ is independent of the amount disorder, even

10 0.4 1 W in the limit of very low WE the same exponent τ = 5/2
8 0.3
0.75 is retained. Note that due to the normalization of the
10 0.55
0.2
distributions along the vertical axis scaling is done with
6 the product of the two exponents α and τ .
0.1
10 We have pointed out in Ref. [22] that when stiffness
0.0
p( )

4 0.0 0.25 0.5 is the only source of disorder Wσ = 0 perfectly brittle

10 behavior occurs for any value of WE . However, in the
2 present study we have shown that adding strength dis-
10 W order leads to stabilization. Avalanche size distributions
1 0.75 are presented in Fig. 6(c) for WE = 1 varying the amount
of strength disorder in a broad range. Again the same
-2 0.55 0.51 functional form occurs as in Fig. 6(a) with the same ex-
10 0.52 0.50 ponent τ = 5/2 for all the parameter sets. The scaling
2 3 4 collapse in Fig. 6(d) is also obtained with the exponent
10 10 10 10 α = 1/2 as for the case of varying stiffness disorder. The
data collapse analysis also implies that the cutoff burst
size ∆c of the distributions has a power law dependence
FIG. 5. (Color online) Burst size distributions p(∆) for sev- on the amount of disorder in the form ∆c ∼ WEα and
eral values of strength disorder Wσ in the absence of stiffness ∆c ∼ Wσα when the stiffness and strength disorder are
disorder WE = 0. Approaching the brittle regime Wσ → 1/2
varied, respectively, in the limit of low disorder.
the well known result of the crossover in τ from 5/2 to 3/2 is
recovered. The two straight lines represent power laws with A very important consequence of the above numeri-
exponents 3/2 and 5/2. The inset presents the constitutive cal analysis is that when approaching the brittle limit
behaviour of the system for three values of Wσ . WE → 0 for Wσ < 1/2 and Wσ → 0 for WE > 0, the
avalanche size distributions do not show any crossover to
a lower exponent. Reducing the amount of disorder both
the number and size of avalanches decreases, however,
fraction of large size events gets higher since the breaking the value of the power law exponent τ remains robust.
of stronger fibers can trigger more secondary breakings. Crossover occurs when solely strength disorder is present
Avalanches of breaking fibers are analogous to acoustic WE = 0.
outbreaks generated by the nucleation and propagation of Note that no similar scaling behaviour can be observed
cracks in heterogeneous solids under an increasing stress when only one source of disorder is present: For con-
[27–29]. Recording the acoustic waves that emanate from stant fiber strength, the system has a perfectly brittle be-
the material is an indicator of the breaking phenomena haviour at any values of WE . When strength is the only
on the microscopic level. It has been addressed that the source of disorder the bundle becomes critical already
crossover in τ could be exploited to forecast the imminent when Wσ approaches 1/2. It has the consequence that
catastrophic failure [23, 30]. in the limit Wσ → 1/2 the characteristic avalanche size
In Sec. III A the analysis of the fraction of broken increases, i.e. just the opposite happens to what we have
fibers Pb at the peak load has already shown that adding presented above. When the two disorders are blended
a second source of disorder to an otherwise brittle sys- the reason of the decreasing avalanche activity is that for
tem, quasi-brittle behavior emerges, i.e. Pb is never zero any WE and Wσ the threshold distribution h(εth ) starts
when Wσ , WE > 0. It has the interesting consequence from a zero value even if the thresholds have a finite lower
that macroscopic failure is always preceded by finite bound εmin
th > 0.
avalanches which behave as precursors to failure. To
demonstrate this effect in Fig. 6(a) avalanche size distri-
butions are presented for Wσ = 0.25 which should pro- VI. DISCUSSION
vide perfectly brittle behavior (no stable avalanches) if
strength is the only source of disorder. Simulations re- The fracture of disordered materials proceeds in bursts
vealed that for any finite value of WE the size distribution which can be recorded in the form of acoustic noise. Mea-
p(∆) has a power law behavior as in Eq. (19) followed by suring crackling noise is the primary source of informa-
a stretched exponential cutoff. For the value of the power tion about the microscopic dynamics and time evolution
law exponent the usual mean field value τ = 5/2 was ob- of fracture. From laboratory experiments through engi-
tained. It can be observed that the amount of disorder neering constructions to the scale of natural catastrophes
WE only controls the total number of avalanches and the forecasting techniques of imminent global failure strongly
cutoff burst size of the distributions. Figure 6(b) shows rely on identifying signatures in the evolving temporal se-
that rescaling the burst size by the α = 1/2 power of WE quence of breaking bursts [31, 32]. When the disorder is
all the distributions can be collapsed on a master curve. absent or its amount is not sufficiently high, failure oc-
This high quality scaling collapse demonstrates that the curs in a catastrophic manner without any precursors.
 8
1
10
-1
(a) -2 (b) trolling the amount of disorder by the width of the dis-
10
tributions. Investigating the microscopic process of fail-
-3 W =0.25

ατ
10 -4

p(∆)WE
10 ure we pointed out that the size of crackling bursts is
p( )

10
-5
10
-6 WE power law distributed followed by an exponential cutoff.
-7 -8 0.1 The power law exponent proved to be equal to the usual
10 10
0.01 0.001 mean field exponent 5/2 without having any crossover
-9 W =0.25 10
-10
10 0.005 0.0005 to a lower value when approaching the limit of perfect
2 3 2 3 4
10 10 10 10 10 10 10 brittleness. The amount of disorder only controls the
α
∆/WE
-1
number of avalanches and their cutoff size. The origin
10
10
-1 (c) (d) of the stabilization mechanism is that the distribution of
-3
10 the relevant failure threshold, obtained as the convolu-
10
-3 p(∆)Wσατ WE=1.0
10
-5
tion of the stiffness and strength distributions of fibers,
p( )

10
-5 -7 W starts from a zero value even if the thresholds have a
10
-7 -9
0.05 finite lower bound.
10 10 0.01 0.001
-9
10
WE=1.0 10
-11
0.005 0.0005
Recently, the problem of mixing strength and stiffness
2 3 2 3 4
10 10 10 10 10 10 10 disorder has been considered in a simplified modelling
∆/Wσα
framework: in Ref. [33] a bundle was composed of a
few groups of fibers of different Young moduli having
FIG. 6. (Color online) (a) Size distribution of bursts for Wσ = uniformly distributed failure strength. Approximate cal-
0.25 varying the stiffness disorder WE in a broad range. (b) culations showed that increasing the number of groups
Scaling collapse of the distributions of (a). Best collapse is of equally spaced Young modulus values the bundle re-
achieved with the exponent α = 1/2. In (a) and (b) the tains the quasi-brittle behaviour for narrower strength
same legend is used. (c) Size distribution of bursts for WE =
distributions. Our analytical results provide a general
1 varying the amount of strength disorder Wσ in a broad
range. (d) Scaling collapse of the distributions of (c) using
understanding of the findings of Ref. [33] with the addi-
the exponent α = 1/2. In (c) and (d) the same legend is used. tional outcome that the continuous stiffness distribution
of our study corresponds to an infinite number of groups
of fibers where stability is ensured for any finite amount
of strength disorder.

Hence, enhancing the quasi-brittle nature of fracture by To test the generality of our results, we also consid-
controlling disorder is of high practical importance. ered the case where the breaking threshold and Young
In the present paper we considered this problem in modulus of fibers follow a Weibull distribution with a
the framework of fiber bundle models. This approach lower cutoff x− . Here x stands for both strength σth and
provides a simple representation of the disorder and al- stiffness E. Simulations performed with several values
lows for the investigation of the microscopic dynamics of of x− verified that any finite amount of disorder leads
the failure process under various types of loading con- to quasi-brittle behavior of the bundle when two sources
ditions. The classical setup of FBMs assumes constant of disorder are present. Furthermore, the burst size dis-
Young modulus of fibers so that the heterogeneous mi- tribution exponent τ displayed a crossover from 5/2 to
crostructure is solely captured by the random strength of 3/2 only when the strength disorder was reduced in the
fibers. Here we proposed an extension of FBMs by con- absence of stiffness disorder in agreement with [34].
sidering simultaneously two sources of disorder, i.e. both
the strength and stiffness of fibers are random variables Besides their theoretical importance our results have
independent of each other. We carried out a detailed ana- practical relevance for materials’ design, the controlled
lytical and numerical investigation of the fracture process blending of stiffness and strength disorder is a promiss-
of the system under quasi-statically increasing external ing way to increase the safety of constructions. Most of
load both on the macro and micro scales. our results are formulated in a general way so that they
For the case of global load sharing we showed that in- can be applied to any strength and stiffness distributions
troducing a second source of disorder stabilizes the sys- used in engineering and materials science. Biological ma-
tem in the sense that a bundle which has a perfectly terials exhibit a broader variety of stiffness and strength
brittle behavior becomes quasi-brittle whenever a finite than engineering materials which could also be captured
amount of disorder of the other field is added. We gave in the framework of our model. An important limitation
a general analytical derivation of the constitutive equa- of our study that has to be resolved is the assumption
tion and of the stability criterion of the system in terms that thrength and stiffness of material elements are un-
of the disorder distributions. For the purpose of nu- correlated. In real materials correlations naturally de-
merical investigations an efficient simulation technique velop such that higher stiffness may be accompanied by
was worked out. As a specific case we considered uni- higher strength. Work is in progress to capture these
formly distributed strength and stiffness of fibers con- types of correlation in our model.
 9

ACKNOWLEDGMENTS

We thank financial support of the projects TAMOP-

4.2.2.A-11/1/KONV-2012-0036, and OTKA K84157.

[1] H. J. Herrmann and S. Roux, eds., Statistical models for [17] K. Okumura and P. de Gennes, Eur. Phys. Jour. E 4, 121
the fracture of disordered media, Random materials and (2001).
processes (Elsevier, Amsterdam, 1990). [18] P. K. Nukala and S. Simunovic, Biomaterials 26, 6087
[2] M. Alava, P. K. Nukala, and S. Zapperi, Adv. Phys. 55, (2005).
349476 (2006). [19] B. E. Layton and A. M. Sastry, Acta Biomaterialia 68,
[3] S. Biswas, P. Ray, and B. K. Chakrabarti, Statisti- 612 (2006).
cal Physics of Fracture, Beakdown, and Earthquake: Ef- [20] F. Barthelat, Phil. Trans. Roy. Soc. A: Math. Phys. Eng.
fects of Disorder and Heterogeneity, Statistical Physics Sci. 365, 2907 (2007).
of Fracture and Breakdown (John Wiley & Sons, New [21] N. Yoshioka, F. Kun, and N. Ito, Phys. Rev. Lett. 101,
York, 2015). 145502 (2008).
[4] O. Ramos, P.-P. Cortet, S. Ciliberto, and L. Vanel, Phys. [22] E. Karpas and F. Kun, Europhys. Lett. 95, 16004 (2011).
Rev. Lett. 110, 165506 (2013). [23] S. Pradhan, A. Hansen, and P. C. Hemmer, Phys. Rev.
[5] Z. Halász, Z. Danku, and F. Kun, Phys. Rev. E 85, Lett. 95, 125501 (2005).
016116 (2012). [24] F. Raischel, F. Kun, and H. J. Herrmann, Phys. Rev. E
[6] J. Vasseur, F. B. Wadsworth, Y. Lavalle, A. F. Bell, I. G. 74, 035104 (2006).
Main, and D. B. Dingwell, Sci. Rep. 5, 13259 (2015). [25] M. Kloster, A. Hansen, and P. C. Hemmer, Phys. Rev.
[7] F. Kun and H. J. Herrmann, J. Mat. Sci. 35, 4685 (2000). E 56, 26152625 (1997).
[8] R. C. Hidalgo, K. Kovács, I. Pagonabarraga, and F. Kun, [26] F. Kun, S. Zapperi, and H. J. Herrmann, Eur. Phys. J.
Europhys. Lett. 81, 54005 (2008). B 17, 269 (2000).
[9] R. C. Hidalgo, F. Kun, K. Kovács, and I. Pagonabarraga, [27] G. Niccolini, A. Carpinteri, G. Lacidogna, and
Phys. Rev. E 80, 051108 (2009). A. Manuello, Phys. Rev. Lett. 106, 108503 (2011).
[10] S. Pradhan, A. Hansen, and B. K. Chakrabarti, Rev. [28] L. I. Salminen, A. I. Tolvanen, and M. J. Alava, Phys.
Mod. Phys. 82, 499 (2010). Rev. Lett. 89, 185503 (2002).
[11] P. V. V. Nukala, S. Simunovic, and S. Zapperi, J. Stat. [29] P. Diodati, F. Marchesoni, and S. Piazza, Phys. Rev.
Mech: Theor. Exp. , P08001 (2004). Lett. 67, 2239 (1991).
[12] G. A. D’Addetta, F. Kun, and E. Ramm, Gran. Matt. [30] S. Pradhan and B. K. Chakrabarti, Phys. Rev. E 65,
4, 77 (2002). 016113 (2002).
[13] H. A. Carmona, F. K. Wittel, F. Kun, and H. J. Her- [31] P. R. Sammonds, P. G. Meredith, and I. G. Main, Nature
rmann, Phys. Rev. E 77, 051302 (2008). 359, 228 (1992).
[14] C. Ergenzinger, R. Seifried, and P. Eberhard, Gran. [32] M. Stojanova, S. Santucci, L. Vanel, and O. Ramos,
Matt. 13, 341 (2010). Phys. Rev. Lett. 112, 115502 (2014).
[15] K. Okumura, Europhys. Lett. 67, 470 (2004). [33] S. Roy and S. Goswami, arXiv:1510.00687 (2015).
[16] C. Urabe and S. Takesue, Phys. Rev. E 82, 016106 [34] S. Pradhan, A. Hansen, and P. C. Hemmer, Phys. Rev.
(2010). E 74, 016122 (2006).