You are on page 1of 10

Physically based non-linear stressstrain relations for the

inter-bre fracture analysis of FRP laminates


A. Puck
a,1
, M. Mannigel
b,
*
a
Am Ahlberg 33, D-34376 Immenhausen, Germany
b
Institute of Plastics Processing, RWTH Aachen University, Germany
Received 10 April 2006; received in revised form 7 October 2006; accepted 8 October 2006
Available online 27 November 2006
Abstract
The high quality of a lamina-by-lamina failure analysis of FRP laminates depends on both realistic failure criteria and a correct
stressstrain analysis. This paper presents some advances in the non-linear stressstrain analysis of UD laminae with non-exibilized
thermoset matrix as they are used in structural applications. In addition to the well-known non-linearity of the stressstrain curve in
pure longitudinal shear s
21
and uniaxial transverse compression r
2
there is an inuence under combined (r
2
, s
21
)-stress of the normal
stress r
2
on the shear strain c
21
and an inuence of the shear stress s
21
on the transverse strain e
2
. Although this strain-interaction is
already known for a rather long period none of the participants of the Worldwide Failure Exercise (WWFE) has included this eect
in the stressstrain analysis. In this paper, a new computational model on a physical basis is presented by which these problems can
be handled very easily. The variable of decisive importance for this procedure is the stress exposure ratio f
E
, which is a measure for
the risk of inter-bre fracture. In the numerical examples f
E
is calculated from the new realistic action plane fracture criteria for brittle
inter-bre fracture.
2006 Elsevier Ltd. All rights reserved.
Keywords: B. Non-linear behaviour; B. Stressstrain curves; C. Failure criterion
1. Introduction
In many FRP components a plane stress combination
r
1
, r
2
, s
21
as shown in Fig. 1 prevails. The lamina-by-lam-
ina strain and fracture analysis for this stress situation has
already reached a high standard. Some shortcomings in the
eld of failure criteria have recently been overcome by the
introduction of the Puck-action plane criteria for inter-bre
fracture (IFF). These are based on a modied Coulomb,
Mohr fracture hypothesis and are formulated with the
three stresses r
n
, s
nt
, s
n1
on the fracture plane, see Fig. 1.
Now the failure criteria are not longer the weakest point
in the theory but there are some shortcomings in the
stressstrain analysis.
In the present work a method for an analytical descrip-
tion of stressstrain relations with (r
1
, r
2
, s
21
) stress states
has been developed. The drop of stiness, which becomes
obvious by an increased attening of the (r
2
, e
2
) and (s
21
,
c
21
) curves as the stress r
2
or s
21
increases is brought into
relationship with the IFF stress exposure ratio f
E
which
also increases with the stress. In this way we can cover ana-
lytically even the observation one frequently encounters
in the literature that a (s
21
, c
21
) curve is inuenced by
the stress r
2
acting simultaneously with s
21
and that con-
versely a stress s
21
acting simultaneously with r
2
will have
an eect on the (r
2
, e
2
) curve. The main reason for this
eect is that the additional stress r
2
or s
21
has an inuence
on the risk of fracture, which can be expressed quantita-
tively by the stress exposure f
E
of the UD lamina. In most
cases, the fracture risk is increased by the additional stress
but with certain ratios of compressive r
2
and s
21
the sus-
tainable shear stress s
21
is, however, increased by the
additional compressive r
2
. In other words, there is less risk
0266-3538/$ - see front matter 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compscitech.2006.10.008
*
Corresponding author. Tel.: +49 241 80 23818; fax: +49 241 80 22316.
E-mail address: mannigel@ikv.rwth-aachen.de (M. Mannigel).
1
Tel.: +49 5673 3517.
www.elsevier.com/locate/compscitech
Composites Science and Technology 67 (2007) 19551964
COMPOSITES
SCIENCE AND
TECHNOLOGY
of fracture when a moderate compressive stress r
2
is added
to a stress s
21
. This is correctly covered by the stress expo-
sure f
E
when this is calculated using realistic inter-bre
fracture criteria. The stress exposure ratio f
E
(or stress
exposure for short) is dened as the length of the vector
{r} representing the acting stress combination divided by
the length of the fracture vector {r}
fr
, which has the same
direction as {r} (that means proportional stretching of all
components of {r} is assumed).
2. The fundamental stressstrain relations of the UD lamina
and their analytical treatment
A survey now follows of the relations normally
employed in the stress analysis of a UD lamina within a
laminate. In Classical Laminate Theory (CLT) linear-elas-
tic behaviour is assumed. When r
1
, r
2
, s
21
are acting simul-
taneously, the strains caused by the single stresses are
simply superimposed. With m
?k
as the larger and m
k?
as
the smaller Poissons ratio then follows:
e
1

r
1
E
k
m
k?
r
2
E
?
1
e
2
m
?k
r
1
E
k

r
2
E
?
2
c
21

s
21
G
?k
3
Resolving Eqs. (1)(3) by the stresses yields:
r
1

E
k
1 m
?k
m
k?
e
1

m
k?
E
k
1 m
?k
m
k?
e
2
4
r
2

m
?k
E
?
1 m
?k
m
k?
e
1

E
?
1 m
?k
m
k?
e
2
5
s
21
G
?k
c
21
6
As can be seen, the elasticity law for the shear stress s
21
(Eq. (6)) is decoupled from the elasticity laws for the nor-
mal stresses r
1
and r
2
(Eqs. (4) and (5)). In other words,
r
1
and r
2
do not inuence the shear strain c
21
, and s
21
does
not inuence the strains e
1
and e
2
. As it will be shown later
in this paper, this complete decoupling does not prove to be
realistic.
The larger Poissons ratio m
?k
is linked to the smaller
ratio m
k?
by the following relation which is obtained when
energy aspects during the loading process are considered:
m
?k
m
k?

E
k
E
?
7
When CLT is applied for a lamina-by-lamina stress and
fracture analysis, the rst step is to calculate the strains
e
1
, e
2
, c
21
which a layer embedded in the laminate experi-
ences, and then after this to calculate the stresses r
1
, r
2
,
s
21
which result from these strains, doing so with the aid
of Eqs. (4)(6). In the case of the fracture analysis of lam-
inates with UD laminae in which high r
2
compressive stres-
ses and s
21
shear stresses occur, considerable errors can
however arise since in reality the stressstrain relations
for these stresses are markedly non-linear. For this reason
a linear calculation often supplies IFF strain limits ,which
are much too low.
As was already pointed out during the WWFE [1], the
procedure often employed to achieve a rst approximation
to reality, although still using the relationships given by
Eqs. (4)(6), nevertheless replaces the modulus of elasticity
E
?
by the secant modulus E
?s
= r
2
/e
2
and correspondingly
replaces the shear modulus G
?k
by the shear secant modu-
lus G
?ks
s
21
=c
21
.
In the fracture analysis of a laminate consisting of UD
laminae, normally stressstrain curves plotted on the
basis of uniaxial r
2
stress or pure s
21
stress are available.
From these, for example, in a lamina-by-lamina stress
analysis in the non-linear domain, one can determine
the secant modulus E
?s
in the case of a transverse com-
pressive stress r
2
. With a UD lamina embedded in the
laminate and loaded by r
2
and r
1
simultaneously, it
would however be wrong to take E
?s
from the uniaxial
(r
2
, e
2
) curve at the strain e
2
which is associated with
the biaxial loading by r
2
and r
1
. This is why it is not
the strain e
2
which is of decisive importance to the micro-
damage state which momentarily obtains but rather the
momentarily acting stress r
2
. But the problem is that r
2
is not known at the beginning. The stress r
2
has, how-
ever, just to be calculated and this should be done using
the correct secant modulus E
?s
. According to Eq. (5) r
2
follows from:
r
2

E
?s
1 m
?k
m
k?
e
2
m
?k
e
1
8
If the strain e
1
on the UD lamina embedded in the laminate
does not happen to be precisely equal to the transverse
contraction e
1
m
k?
e
2
corresponding to a uniaxial load-
ing by r
2
, then e
1
will exert an inuence on the stress r
2
which arises, as can be seen from Eq. (8). Since m
?k
is the
larger of the two Poissons ratios, at relatively high values
of e
1
and relatively small values of e
2
it can very easily be
the case that the term e
2
m
?k
e
1
has a dierent sign from
e
2
. This means that when e
2
< 0 a tensile stress r
2
occurs
Fig. 1. Stresses which act on planes vertical to the natural axes x
1
, x
2
,
and stresses r
n
, s
nt
, s
n1
acting on a parallel-to-bre fracture plane inclined
by the angle h
fp
against the thickness direction x
3
.
1956 A. Puck, M. Mannigel / Composites Science and Technology 67 (2007) 19551964
and when e
2
> 0 a compressive stress r
2
occurs. In a previ-
ous paper we examined in detail the errors which can arise
[2].
As has been indicated above, the drop in stiness,
expressed by E
?s
< E
?
, is connected mainly with the micro-
damage, and the latter depends before anything else on
the stress state and not on the strain e
2
. More recent com-
puting programs have for this reason been designed such
that for a given momentary (e
1
, e
2
, c
21
) strain state, the
momentary stress r
2
is calculated using a secant modulus
E
?s
which is tapped o from the uniaxially plotted (r
2
,
e
2
) curve, see [3], but in which the e-axis now represents
the combined strain
e
comb
2;1

e
2
m
?k
e
1
1 m
?k
m
k?
: 9
In other words, that although r
2
is formally calculated in
the same way as a uniaxial stress state by Hooks law
r = E e, at the same time the interaction of e
1
and e
2
in
bringing about the stress r
2
is however correctly described
by:
r
2
E
?s
e
comb
2;1
10
The combined strain e
comb
2;1
is the magnitude of the strain
which in the case of uniaxial loading would result in the
same stress r
2
as the simultaneously occurring strains e
1
and e
2
create in the present biaxial case. Equal stress r
2
means that in both cases the same secant modulus E
?s
ap-
plies. If the (r
2
, e
2
) curve gained in a uniaxial test has been
described mathematically by a function r
2
= F
r
e
2
, then in
the biaxial case the stress r
2
can be calculated by
r
2
F
r
e
comb
2;1
and the corresponding secant modulus E
?s
then follows from E
?s
r
2
=e
comb
2;1
. Using e
comb
2;1
in a stress
and fracture analysis at the right time can improve the con-
vergence behaviour of computing programs. This is ex-
plained in more detail in [3].
One question still remains open as to whether and if
applicable, how Eq. (7) should be used in the case of
non-linear calculations. Eq. (7), which applies to linear-
elastic behaviour, derives from an energy balance of a
(r
1
, r
2
)-stress state in which in one case r
1
must be applied
rst but in which in the other case r
2
must be applied rst.
In the nal state, the same deformation energy must have
been stored in the material element, irrespective of the
sequence of loading. A corresponding approach will in
principle apply even in the case of non-linear behaviour
and it should also lead to a similar result despite the fact
that the damage-related lost energy now also plays a role
in the energy balance. For this reason even when calcula-
tion is performed non-linearly, the relation expressed in
Eq. (7) between the Poissons ratios will be kept but with
E
?
being replaced by E
?s
.
As stress exposure increases, micromechanical consider-
ations speak in favour of also reducing m
?k
in the same way
as E
?
is reduced to E
?s
[4]. However, this has not yet been
conrmed by measurements [5]. To avoid excessively low
transverse stresses r
2
being calculated in the case of safety
verications (cf. Eqs. (5), (8) and (10)), calculations should
as a precautionary measure be carried out using a xed
value for m
?k
.
A fundamental weakness in the non-linear procedure of
calculation presented here is that it does not take into con-
sideration the inuence which we have mentioned above
of r
2
exerted on the (s
21
, c
21
) curve and of s
21
exerted on
the (r
2
, e
2
) curve.
3. A new computational model for the analytic description of
non-linear stressstrain relations under simple and combined
loading by r
1
, r
2
, s
21
3.1. Theoretical development of the model
Due to the irregular microstructure of the bre/matrix
composites and the chaotic microdamage processes which
take place there, presumably there will be virtually no
chance of describing analytically the loss in stiness in
the UD lamina which increases with loading by microme-
chanical approaches. Due to the known global correlation
between a progressive loss of stiness in the UD lamina
and the stress exposure f
E
a numerical value which indi-
cates the level of the fracture risk for IFF and the micro-
damage associated with this the possibility does however
exist of developing a macroscopic model which will be ade-
quate for the purposes of laminate fracture analysis within
the context of component development.
Since non-linearity is particularly marked in the case of
loading by longitudinal shear s
21
and some informative
experimental ndings are already available concerning
the eects of r
2
compressive stress on shear strain c
21
, let
us begin with an approach applicable to pure s
21
shear
stress. As we know from experience, distinct microdamage
and the non-linearity resulting from this does not occur
until the stress exposure exceeds a particular threshold
value, such as f
s
21

Ethr
. Accordingly, the following approach
is taken as regards the longitudinal shear secant modulus
G
?ks
s
21
=c
21
:
G
?ks
G
?k

f
s
21

E
f
s
21
Ethr
1 f
s
21

Ethr
_ _
n
s
21

G
?k
G
?ks
j
f
s
21

E
1

11
In what follows, the value f
s
21

E
is used to describe the stress
exposure which arises from the application of pure s
21
.
With the transverse/longitudinal shear strength R
?k
it is
computed as f
s
21

E
js
21
j=R
?k
. The linear range of the
(s
21
, c
21
) curve is limited by f
s
21

Ethr
. In the case of pure shear
s
21
, IFF occurs when f
s
21

E
becomes equal to 1. As IFF oc-
curs at f
s
21

E
1, the dierence between the secant modulus
G
?ks
and the initial modulus G
?k
is at its greatest. When
there is a gradual increase in loading which goes beyond
the threshold value f
s
21

Ethr
, the dierence between the
momentary G
?ks
and G
?k
increases progressively with
f
s
21

E
. Mathematically this is already considered by using
A. Puck, M. Mannigel / Composites Science and Technology 67 (2007) 19551964 1957
the dierence f
s
21

E
f
s
21

Ethr
. This eect is further intensi-
ed by an exponent n
s
21

> 1. The variables f


s
21

Ethr
and
n
s
21

can be calculated from experimentally obtained (s


21
,
c
21
) curves by numerical curve tting.
As regards the bre/matrix systems used in the WWFE,
the (s
21
, c
21
) curves at rst loading have been documented
up to IFF by specifying 20 and more (s
21
, c
21
) values [6].
We have evaluated this information for an GFRP system
and for a CFRP system by making a curve tting of the
parameters f
s
21

Ethr
and n
s
21

to the given values of (s


21
, c
21
).
We have to keep in mind that Eq. (11) is an approxima-
tion of the non-linear part of the stressstrain curve
s
21
G
?ks
c
21
. Therefore only values (s
21
, c
21
) which
belong to the non-linear part of the (s
21
, c
21
) curve can
be used for the curve tting. The missing linear part is given
by s
21
G
?k
c
21
. The result is shown in Fig. 2.
The curve tting parameters turned out to be for the
E-glass/LY750/HY917/DY063 system at f
s
21

E
0:48 and
n
s
21

1:35, and for the AS4/3501-6 system at


f
s
21

E
0:22 and n
s
21

1:90. In the same way as in


Fig. 2, several other examples also showed the selected
model based on Eq. (11) can deliver a very good approxi-
mation to various measured (s
21
, c
21
) curves.
If additional stress r
2
is superimposed to the s
21
-stress,
the (s
21
, c
21
) curves are not fundamentally changed they
still retain their essential shape. In most cases the curves
are somewhat atter now than the curve for pure s
21
shear.
However, what is at rst sight surprising is the fact that a
moderate transverse compressive r
2
stress lifts the (s
21
,
c
21
) curve a little and also extends it to higher values of
the fracture shear strain c
21
at IFF. But, there is a simple
explanation for this: transverse compressive stresses r
2
act-
ing simultaneously with s
21
in a specic domain of the ratio
r
2
/s
21
result, when fracture under the (r
2
/s
21
) combination
occurs, in values being reached for js
21
j which are greater
than the shear strength R
?k
. This strength increase can
amount to 30% or so, as it derives from the action plane
fracture criterion for IFF of Mode C. It should be noted
that the maximum attainable shear stress s
21max
occurs at
r
2
0; 5R
c
?
(see also Fig. 4b) and amounts to:
js
21max
j R
?k
1 p
c
??
12
where p
??
is a so-called inclination parameter with a
magnitude of 0.20.3. In the case of pure s
21
stress,
f
s
21

E
in Eq. (11) is the variable which controls the drop
in stiness. If s
21
and r
2
act simultaneously, it will now
obviously depend on how much the total stress exposure
f
E
caused by the combined (r
2
, s
21
) loading is greater or
smaller than the stress exposure f
s
21

E
caused by s
21
alone.
Accordingly, to the numerator at f
s
21

E
we add the dier-
ence f
E
f
s
21

E
which, to allow adaptation to experimen-
tal ndings, gets a pre-factor C
r
2

. The equation for the


shear secant modulus G
?ks
with combined (r
2
, s
21
) stress
is thus:
G
?ks
G
?k

f
s
21

E
C
r
2

f
E
f
s
21

E
f
s
21

Ethr
1 f
s
21

Ethr
_ _
n
s
21

G
?k
G
?ks
j
f
s
21

E
1
13
The letter C (combined) has been chosen for the pre-fac-
tor, because it controls the eect of combined stress, and
the superscript (r
2
) relates the pre-factor C to the eect
of an additional stress r
2
. (Because the eect of an addi-
tional tensile r
2
may dier from the eect of a compressive
r
2
it could prove useful in some cases to use dierent super-
scripts, r
t
2
or r
c
2
).
The eect of the dierence C
r
2

f
E
f
s
21

E
which has
been introduced for the inuence of the additional stress
r
2
on the (s
21
, c
21
) curve is only relatively small, as it is also
found experimentally. (In a computing program Eq. (13)
can of course also be applied to the case of pure s
21
stress
state since the dierence f
E
f
s
21

E
in this case becomes
zero.)
Currently, the state of the art in describing the eect of
r
2
on the (s
21
, c
21
) curve is, that based for instance on some
measured (s
21
, c
21
) curves for dierent ratios of (r
2
/js
21
j) a
special mathematical surface G
?ks
f e
2
; c
21
[4] or
G
?ks
f r
2
; s
21
[7] is described. This surface is created
by using a pure mathematical not a physically based
interpolation procedure. Our physically based approach
according to Eq. (13) has the advantage that fewer experi-
mental measuring points are required for calibration. In
addition it is easier to become aware of any measurement
errors there may be.
For adaptation to experimentally determined curves in
the case of (r
2
, s
21
) stress combinations, a total of three
parameters is now available in addition to the measured
variables of initial modulus G
?k
and secant modulus
G
?ks
j
f
s
21
E
1
at IFF under pure shear s
21
. These parameters
are f
s
21

Ethr
, n
s
21

and C
r
2

. In our view a sensible way of


determining them is as follows. First of all, f
s
21

Ethr
and
n
s
21

are gained by curve tting for the case of pure s


21
loading. Provided that a few measurement curves are avail-
able for (r
2
, s
21
) stress combinations for a r
2
/js
21
j ratio
which for each curve remains constant, the next step, to
achieve a better t of the curve, is to optimize only the coef-
Fig. 2. (s
21
, c
21
) curves for CFRP and GFRP resulting from pure s
21
stress.
1958 A. Puck, M. Mannigel / Composites Science and Technology 67 (2007) 19551964
cient C
r
2

while retaining the values for f


s
21

Ethr
and n
s
21

previously found for pure shear s


21
.
Analogously to Eq. (13), the equation for determining
the secant modulus E
?s
= r
2
/e
2
for a transverse stress r
2
in combination with a shear stress s
21
is as follows:
E
?s
E
?

f
r
2

E
C
s
21

f
E
f
r
2

E
f
r
2

Ethr
1 f
r
2

Ethr
_ _
n
r
2

E
?
E
?s
j
f
r
2

E
1
14
The value f
r
2

E
denotes the stress exposure resulting from a
uniaxial transverse stress r
2
. In the case of transverse ten-
sion, this will be f
r
2

E
r
2
=R
t
?
, while with transverse com-
pression f
r
2

E
r
2
=R
c
?
applies.
Using Eq. (14) with f
E
f
r
2

E
0 we have also mod-
elled the (r
2
, e
2
) curves resulting from uniaxial r
2
-stress
for the bre/epoxy systems already treated in Fig. 2 (see
Fig. 3).
Here too it appears that by choosing the parameters
f
r
c
2

Ethr
and n
r
2

the model can once again be matched very


well to non-linear material behaviour. For the E-glass/
LY750/HY917/DY063 system we obtain f
r
c
2

Ethr
0:48 and
n
r
c
2

2:36, and for the AS4/3501-6 system f


r
c
2

Ethr
0:45
and n
r
c
2

2:85.
The validity of Eqs. (13) and (14) starts when the stress-
dependent stress exposure term exceeds the threshold value
f
s
21

Ethr
or f
r
2

Ethr
and nishes upon occurrence of IFF in other
words, when the stress exposure taking all stresses into
account becomes f
E
1.
In the course of an iteration process the upper limit of
validity, f
E
= 1, may be of course exceeded temporarily.
The Eqs. (13) and (14) are able to extrapolate the stress
strain curve.
Due to premature tearing of individual elemental bres
or smaller bundles of bres or occasional cases of local
kinking, also high longitudinal stresses r
1
can cause stiness
reducing microdamage in the matrix and at the bre/matrix
interface. This microdamage also weakens the ability of the
UD lamina to resist IFF. In fracture analysis this is taken
into account by a weakening factor g
w1
dependent on r
1
,
see Annex 1.2 of [3]. With this factor the strengths R
t
?
; R
c
?
,
R
?k
which are of decisive importance to IFF are decreased
to reduced values g
w1
R
t
?
, g
w1
R
c
?
; g
w1
R
?k
[3,13]. By imple-
menting a reduction of this kind in the strengths, the IFF
stress exposure f
E
caused by a particular (r
2
, s
21
) stress state
is driven up to a higher level. What this means in the case of
the model based on Eqs. (13) and (14) is that instead of hav-
ing to calculate using f
E
as previously, f
E
0
=g
w1
should now
be used. Here f
E
0
is the IFF stress exposure which has been
calculated without r
1
inuence. (The formula for g
w1
is
given in the Appendix (Eq. (A.6)) as is more detailed infor-
mation in this regard.)
The f
E
dependent calculation of a secant modulus with
the aid of Eqs. (13) and (14) ts very harmoniously into
the usual iterative calculation used in lamina-by-lamina
fracture analysis [3] since the stress exposure f
E
in the indi-
vidual layers of the laminate is calculated at every load step
anyway. All that is now needed in addition is
f
s
21

E
js
21
j=R
?k
and f
r
2

E
r
2
=R
t
?
or f
r
2

E
r
2
=R
c
?
.
After IFF has occurred, the decrease in stiness (degrada-
tion) has already for a long time been described with the
aid of the stress exposure f
E
[3,5,8].
In the computing program, iteration has to be continued
until the calculated strains e
1
, e
2
, c
21
and the stress expo-
sures f
E
, f
r
2

E
, f
s
21

E
have been suciently harmonized with
each other in other words, change only to a very minor
extent with each successive iteration.
3.2. Visualization of the stress exposure ratios involved in the
analytical model
Fig. 4 shows some relations between the stress exposures
f
E
, f
s
21

E
and f
r
2

E
in Eqs. (13) and (14).
Speaking quite generally, stress exposure ratio (or stress
exposure for short) is the ratio of the length of a stress vec-
tor and the length of the corresponding fracture vector.
What corresponding means is that the fracture vector
has the same direction as the vector of the stresses present
which is under consideration. To put it another way, in
both stress states the individual components of the stress
vector stand in the same relation to each other.
In Fig. 4 there are only two components, namely r
2
and
s
21
. In each of the four parts (ad) of the gure a fracture
vector is marked which consists of {r
2
} and {s
21
} and each
of these fracture vectors touches the (r
2
, s
21
) fracture curve
at a dierent place. This curve is assembled from three sec-
tions, each of which is assigned to one particular fracture
mode A, B or C (see Fig. 4a and Appendix).
In Mode A and Mode B the fracture angle h
fp
is h
fp
=
0. In Mode C oblique fracture planes occur, that means
h
fp
50. Vectors of the existing stresses are represented
by thick single line arrows and the corresponding failure
vectors by thin double line arrows. In the case of the vector
{r}
fr
resulting from {r
2
} and {s
21
}, the single line and the
double line arrows are of equal length in all parts (ad) of
the gure, since f
E
= 1. On the r
2
axis and the s
21
axis the
corresponding single line arrows represent the relevant
component {r
2
} and {s
21
} of the fracture vector {r}
fr
and the corresponding fracture vectors are also drawn as Fig. 3. r
c
2
; e
2
curves resulting from uniaxial r
2
stress.
A. Puck, M. Mannigel / Composites Science and Technology 67 (2007) 19551964 1959
a double line vector for uniaxial transverse stress r
2
and
pure shear s
21
. Their length is equal to R
t
?
or R
c
?
or R
?k
,
respectively.
In Fig. 4a it can be seen that in Mode A, when f
E
= 1,
then not only is f
r
2

E
< 1 but also f
s
21

E
< 1. In Fig. 4b
the fracture vector touches the (s
21
, r
2
) curve at its highest
point. On the s
21
axis it can be seen that now the corre-
sponding value f
s
21

E
> 1. This makes the dierence
f
E
f
s
21

E
in Eq. (13) negative. On the other hand,
f
r
2

E
< 1, namely f
r
2

E
0:5. In Fig. 4c not only is f
E
= 1
but also f
s
21

E
1. As a result the dierence which controls
the inuence of r
2
on the (s
21
, c
21
) curve becomes
f
E
f
s
21

E
0. Therefore, according to Eq. (13) for this
(r
2
, s
21
) combination we obtain the same (s
21
, c
21
) curve
as for pure s
21
shear. In contrast to this, the shown stress
combination leaves f
r
2

E
< 1. In Fig. 4d, where jr
2
j is con-
siderably greater than s
21
, not only is f
r
2

E
< 1 but also
f
s
21

E
< 1. While the dierence f
E
f
s
21

E
which appears
in Eq. (13) can be negative in a particular section of the
fracture curve, the dierence f
E
f
r
2

E
in Eq. (14) will
clearly always have positive values.
On the basis of these reections we can already come to
some fundamental understanding regarding the course of
the stressstrain curves to be plotted using Eqs. (13) or
(14), respectively.
3.3. Application of the model to available experimental
results
At the Institute of Plastics Processing (IKV) in Aachen,
Kopp made experimental measurements yielding (s
21
, c
21
)
curves on the basis of a CFRP system (T300/LY556/
HY917/DY070) under combined loading [4] (see Fig. 5).
A rst qualitative judgment shows that the model proposal
described, closely corresponds to reality. What actually
happens is that the course of the (s
21
, c
21
) curve is shifted
upward by a moderate superimposed transverse compres-
sive stress while on the other hand superimposed transverse
tensile stresses or high transverse compressive stresses
result in reduced shear stresses for given shear strains.
The (s
21
, c
21
) curves calculated on the basis of Eq. (13)
for dierent ratios of r
2
/s
21
are shown in Fig. 6. The calcu-
lations are based on the stressstrain curves of the material
system shown in Fig. 5 [4]. For the parameters we obtain
the values f
s
21

Ethr
0:1 and n
s
21

1:59. Optimization of
the parameter C
r
2

was carried out with a view to get the


best possible matching of the experimentally obtained
secant moduli G
?ks
at IFF (f
E
= 1) with transverse stresses
Fig. 4. Stress exposures f
E
= 1 and also the corresponding stress exposures f
s
21

E
and f
r
2

E
demonstrated with the aid of a fracture vector {r}
fr
and its
components {r
2
} and {s
21
}.
Fig. 5. Measured (s
21
, c
21
) curves for combined loading with a CFRP/
epoxy system (T300/LY556/HY917/DY070) [4].
1960 A. Puck, M. Mannigel / Composites Science and Technology 67 (2007) 19551964
r
2
superimposed at dierent levels. In the case of the bre/
matrix system selected, optimization results in C
r
2

0:74.
It can be seen that by our model the (s
21
, c
21
) curve is in
some cases shifted to higher s
21
values and in others to
lower s
21
values, depending on the ratio r
2
/s
21
. As dis-
cussed before this is correct. With our model, the measured
(s
21
, c
21
) curves from [4] as a whole are reproduced satisfac-
torily in the case of the selected parameters. In Fig. 6 there
are three curves for a negative stress ratio r
2
/js
21
j and one
curve for a positive ratio of 1. Preliminary, we have used
the same parameter C
r
2

in both cases, that means inde-


pendently from the fracture mode (r
n
P0 or r
n
< 0).
Whether this is acceptable or not should be investigated
by further theoretical and experimental work.
For the present bre/matrix system, a ratio of
r
2
=js
21
j R
A
??
=R
?k
0:74 delivers the maximum
upward shift of the curve. In this case the value f
s
21

E
is con-
siderably higher than the total stress exposure f
E
which is
f
E
= 1 (cf. discussion of Fig. 4b).
Fig. 7 shows (r
2
, e
2
) curves calculated with Eq. (14) for
dierent ratios of js
21
j/r
2
. In the tests carried out by Kopp
the experimental results, which in Fig. 7 too are plotted
only as end points of the curve in question, were always
obtained simultaneously with the corresponding (s
21
, c
21
)
curve and in all cases using one and the same test specimen.
In the present case, numerical adaptation of the param-
eters results in f
r
2

Ethr
0:05, n
r
2

2:18 and C
s
21

i
0:53.
If a compressive stress r
2
is superimposed by a shear
stress s
21
, this, as compared with a uniaxial transverse com-
pressive stress, will always lead to an increase in the stress
exposure f
E
(cf. Fig. 4), the sign of the shear stress s
21
has
no inuence on this. This causes the (r
2
, e
2
) curves to be all
shifted upwards in other words, towards lower compres-
sive stresses. The (r
2
, e
2
) curves given by Kopp in [4] are
reproduced quite satisfactorily by Eq. (14). Based on the
experimental data of Kopp the mathematic curve tting
program provided unusual low values for f
s
21

Ethr
and f
r
2

Ethr
.
These are accompanied by rather high values for n
s
21

and n
r
2

. Obviously the eect of a low magnitude of f


Ethr
is more or less compensated by a high exponent n. Later
on some general recommendations for suitable values of
the parameters are given, making use of this nding.
Within the tension range of r
2
the (r
2
, e
2
) stressstrain
curves are almost straight-line curves in other words,
there is E
?s
= E
?
everywhere. This means that even on
the basis of Eq. (14), s
21
cannot gain an inuence on the
(r
2
, e
2
) diagram, because the decisive factor at the end of
the formula is E
?
E
?s
j
f
r
2

E
1
0. However, if it is
wished to bring an inuence of this kind into eect in the
calculation, this will require taking E
?s
at IFF as somewhat
smaller than E
?
or otherwise the last factor in Eq. (14) will
always remain equal to zero. In this paper we will not deal
with this task, because at the moment we do not see any
necessity.
Fig. 8 shows results from the earliest work [9] of which
we are aware and which deals with the mutual inuencing
of stressstrain curves of r
2
and s
21
. These results were pro-
cessed on the basis of our computational model, Eq. (13).
The work [9] was concerned with, among other things,
investigating experimentally the inuence of stresses r
2
on the (s
21
, c
21
) curve for a bre/matrix system frequently
used in practice, namely E-glass/epoxy (LY556/HY917/
Fig. 6. (s
21
, c
21
) curves calculated on the basis of Eq. (13), with
superimposed r
2
stress. Only the end points of the corresponding
experimentally determined curves for T300/LY556/HY917/DY070 [4]
are plotted, compare Fig. 5.
Fig. 7. (r
2
, e
2
) curves calculated on the basis of Eq. (14), with
superimposed s
21
shear stress and showing end points of the correspond-
ing experimentally obtained curves [4].
Fig. 8. (s
21
, c
21
) curves calculated with Eq. (13) in the case of superim-
posed r
2
stress. The end points of the corresponding experimentally
determined curves for E-glass/LY556/HY917/DY070 are plotted [9].
A. Puck, M. Mannigel / Composites Science and Technology 67 (2007) 19551964 1961
DY070). The loading of the specimens was accompanied
with the monitoring of acoustic emission. Depending on
the magnitude of the superimposed transverse stress r
2
the acoustic emission starts between shear stresses from
s
21
= 10 to 30 N/mm
2
, that means in the region of
f
s
21

Ethr
% 0:15 0:4.
Fig. 8 shows the results of our evaluation carried out
with the aid of Eq. (13) and here too the end points of
the (s
21
, c
21
) curves obtained experimentally by Schro der
have been plotted. In the present case, optimization of
the parameters results in f
s
21

Ethr
0:27, n
s
21

2:08 and
C
r
2

0:49. The course of the curve for pure s


21
very clo-
sely resembles that for E-glass/LY750 in Fig. 2. In Fig. 8
f
s
21

Ethr
is somewhat lower than in Fig. 2 but this is compen-
sated by a value of n
s
21

which is higher than in Fig. 2.


Swanson et al. [10] have also carried out fracture tests
on test specimens made of CFRP (AS4/55A) with com-
bined (r
2
, s
21
) stress and plotted the corresponding (s
21
,
c
21
) curves. In these tests, however, s
21
and r
2
were not
applied simultaneously; instead, r
2
was applied rst to
its full level and s
21
only afterwards. It was during this
second loading process with s
21
that the (s
21
, c
21
) curve
was plotted. The stress s
21
which increased continually
until IFF occurred was therefore superimposed right from
the start on a preloading r
2
in a similar way to when a
residual stress r
2
occurs in a laminate. The (s
21
, c
21
)
curves calculated under this precondition using Eq. (13)
will therefore be expected, at least in their initial sections,
to dier more or less markedly from those obtained for
loading processes in which r
2
and s
21
remain permanently
in the same ratio to each other. The result of our calcula-
tion is shown in Fig. 9.
In the present case, curve tting of the parameters
results in f
s
21

Ethr
0:18, n
s
21

2:4 and C
r
2

i
0:58. The
model satisfactorily reproduces the measured values, even
when there is a superimposed constant transverse stress
r
2
. However, the applied r
2
values are still a fair distance
from the corresponding strengths R
t
?
26:7 N=mm
2
and
R
c
?
94:7 N=mm
2
.
4. Literature research
In literature we are aware of the existence of only few
papers dealing with the subject-matter with which we are
concerned here. In a recent paper Camanho et al. [11]
quoted the following purely empirical formula (originating
from Hahn and Tsai) for describing the (s
21
, c
21
) curve,
namely
c
21

s
21
G
?k
b s
3
21
15
and they expanded it with an empirical supplement in this
way:
c
21

s
21
G
?k
b
0

r
2
v
_ _
s
3
21
for r
2
> 0 16
Swanson et al. [10] cite a formulation by Hashin et al. from
1974 which resembles our formulation at least in that not
only momentary stresses s
21
and r
2
but also something sim-
ilar to strength, in this case yield stresses R
?ky
and R
?y
are
included in an equation from which G
?ks
can be calculated:
G
?ks

G
?k
1
s
21
R
?ky
_ _
2

r
2
R
?y
_ _
2
_ _
N1=2
17
The constant N which appears in the exponent (for exam-
ple N = 3.6) is used for curve tting. As even Swanson
et al. [10] complain, Eq. (17) does not react to the sign of
r
2
in the way that our experimentally validated experience
requires. The Eq. (17) is based on the Mises yield criterion
which clearly is not appropriate to the physics of the bre/
matrix composite.
5. Outlook and recommendations
It is expected that in design practice very often reliable
values for f
Ethr
, n, C are not available for the bre/matrix
system used for the component, which has to be designed
and dimensioned. In such cases, where reliable data is miss-
ing, there seems to be a tendency in practice to stay with a
linear-elastic calculation. Very often this leads to uneco-
nomic and too heavy components in cases, where a rela-
tively high margin of safety against IFF is required by
regulations issued by the authorities.
Certainly this paper will stimulate further tests on impor-
tant bre/matrix systems in order to get more reliable results
for f
Ethr
, n and C. In order to systematize future experimen-
tal investigations it ought to be advantageous to select for
the stress ratios not smooth gures as previously done
but rather to standardize the experiments on the basis of
the corresponding strengths. In this case not r
2
/js
21
j = 1
but rather r
2
=js
21
j R
t
?
=R
?k
would be selected and, for
example, correspondingly in the range of the highest sustain-
able s
21
-stress r
2
=js
21
j R
A
??
=R
?k
. An other interesting
ratio could be r
2
=js
21
j R
c
?
=R
?k
. In this way the results
for dierent systems would become more comparable.
Fig. 9. (s
21
, c
21
) curves calculated using Eq. (13) with superimposed r
2
stress of a constant magnitude and test specimens made of AS4/55A. The
curve end points from the tests of Swanson et al. [10] are shown.
1962 A. Puck, M. Mannigel / Composites Science and Technology 67 (2007) 19551964
Of course additional experiments are very desirable.
Nevertheless, we feel already able to present some typical
values for f
Ethr
, n and C. It seems to be dicult to x a pre-
cise value of f
Ethr
from a measured stressstrain curve. As
already mentioned a low value of f
Ethr
can be compensated
by a rather high value of n and vice versa a rather high f
Ethr
by a lower n. Therefore, we propose to use f
Ethr
= 0.3
throughout.
We have recalculated all former examples now starting
with a xed f
s
21

Ethr
0:3 or f
r
2

Ethr
0:3 in order to nd the
corresponding optimal n
s
21

or n
r
2

, respectively. As the
outcome of this procedure some preliminary parameters
which we feel can be recommended for most practical
design purposes are presented. These values for f
Ethr
, n
and C are listed in Table 1.
Especially the value of the parameter C should be exam-
ined by further experiments in future. In any case the use of
the proposed model with parameters from Table 1 will oer
a great advantage compared to a linear-elastic analysis.
6. Concluding remarks
The accuracy of a lamina-by-lamina fracture analysis of
a laminate depends both on the fracture criteria used and
on the method of non-linear stressstrain analysis. A new
computational model for non-linear stressstrain analysis
is presented for the (r
1
, r
2
, s
21
) stress states of UD laminae
which are predominantly found in practice. For combined
stresses the variable of decisive importance is the stress
exposure f
E
. In the inter-bre fracture analysis the values
for E
?s
and G
?ks
are selected with the aid of the momentary
values of the stress exposures f
E
; f
r
2

E
; f
s
21

E
. The model
even covers the inuence of r
2
on the (s
21
, c
21
) curve and
the inuence of s
21
on the (r
2
, e
2
) curve. In this way the
unrealistic decoupling of r
1
, r
2
on the one hand and of
s
21
on the other hand which has been usual until now is
overcome. Comparisons of the results of analyses and
experimentally obtained stressstrain diagrams conrm
the usefulness of the model. What should be particularly
emphasized is the fact that a superimposed r
2
compressive
stress which has a lifting or lowering eect on the (s
21
, c
21
)
curve, depending on the stress ratio r
2
/js
21
j, is correctly
reproduced by the model. This means that now the Puck-
Theory of failure of laminates [3,1214] is a self-contained
lamina-by-lamina fracture theory in which the stress expo-
sure f
E
is the decisive variable controlling the course of
computation throughout from the start of loading up
until the occurrence of IFF and FF. A very comprehensive
description of the features of Pucks action plane fracture
criteria can be found in [14].
A calculation software, which considers the non-linear
behaviour of the lamina and the Puck theory of failure
for the prediction of the strength of laminates as described
in this paper is under development at the Department
Konstruktiver Leichtbau und Bauweisen (KLuB), headed
by Prof. Schu rmann, of Darmstadt Technical University.
The program is named AlfaLam.nl, (Advanced layerwise
failure analysis of Laminates.non-linear). The program is
based on MS EXCEL and can be downloaded from the
homepage of KLub (http://www.klub.tu-darmstadt.de).
The actual direct link to KLub-downloads is: http://
www.klub.tu-darmstadt.de/forschung/download.php.
Appendix A
The examples of the application of the non-linear stress
strain analysis presented in this paper have been calculated
using the analytical solutions for the stress exposure ratio
f
E
which have been published rst in [13]. They can also
be found in [3]. Using these convenient analytical solutions
requires, that the following coupling of parameters is
respected:
p
c
?k
R
?k

p
c
??
R
A
??
A:1
In consequence of this parameter-coupling R
A
??
has to be
calculated from R
?k
and p
c
?k
by the following equation:
R
A
??

R
?k
2p
c
?k

1 2p
c
?k
R
c
?
R
?k

1
_ _
A:2
The necessary equations for the calculation of the stress
exposure f
E
and their limits of validity are to be found in
Table A.1.
In Fig. A.1 the predicted fracture curve based on these
equations is shown. The recommended values for the incli-
nation parameters can be seen in Table A.2. They are dif-
ferent for glass- and for carbon bre reinforced plastics.
The weakening factor g
w1
which includes the eect of r
1
on the IFF strengths R
t
?
, R
c
?
, R
?k
is calculated using the fol-
lowing equation:
g
w
1

ca

c
2
a
2
s
2
1
_
s
ca
2
1
where c
f
E
0
f
EFF
and a
1 s

1 m
2
p A:6
The equation is valid for
1
s
Pc Pm. The stress exposures
calculated for Modes A, B and C must be increased by
dividing f
E
0
by g
w1
. In the formula (A.6) for the weakening
factor f
E
0
means the IFF stress exposure without r
1
inu-
ence. f
EFF
r
1
=R
t
k
for r
1
> 0 or f
EFF
r
1
=R
c
k
for
r
1
> 0 is the bre-fracture stress exposure. sR
t
k
or sR
c
k

is the stress r
1
at which weakening due to r
1
begins to be-
come noticeable. The remaining minimum value of the IFF
strengths in the form of mR
t
?
; mR
c
?
or mR
?k
are given by the
Table 1
Recommended preliminary values of the parameters of the model
Load case f
Ethr
n C
CFRP Shear s
21
0.3 1.7 0.6
Transverse compression r
2
0.3 2.5 0.6
GFRP Shear s
21
0.3 2.0 0.6
Transverse compression r
2
0.3 3.0 0.6
A. Puck, M. Mannigel / Composites Science and Technology 67 (2007) 19551964 1963
fraction m which applies when r
1
is theoretically equal to
R
t
k
or R
c
k
. In the absence of test results it is recom-
mended that calculations use s = m = 0.5. In principle dif-
ferent values of (s, m) can be used for r
1
> 0 and r
1
< 0.
References
[1] Hinton MJ, Kaddour AS, Soden PD. Failure criteria in bre
reinforced polymer composites: the world-wide failure exercise.
Amsterdam: Elsevier; 2004.
[2] Puck A. Should bre-plastics composites be designed with strain or
stress criteria? Kunsts - Ger Plast 1992;82(5):348 [German text pp.
431-434].
[3] VDI Guideline 2014 Part 3. Development of bre-reinforced plastics
components, analysis (bilingual, German and English).Berlin: Beuth-
Verlag, 2006.
[4] Kopp JW. Zur Spannungs-und Festigkeitsanalyse von unidirektion-
alen Faserverbundkunststoen (Contributions to the stress and
strength analysis of unidirectionally bre reinforced plastics). PhD
Thesis, RWTH Aachen University, ISBN 3-89653-438-6; 2000 [in
German].
[5] Knops M. Sukzessives Bruchgeschehen in Faserverbundlaminaten
(Gradual failure process in bre/polymer laminates). PhD Thesis,
RWTH Aachen University, ISBN 3-86130-480-5; 2003 [in German].
[6] Soden PD, Hinton MJ, Kaddour AS. Lamina properties, lay-up
congurations and loading conditions for a range of bre reinforced
composite laminates. Compos Sci Technol 1998;58(7):101122.
[7] Weber T, Schu rmann H. Verbesserung der nichtlinearen Versa-
gensanalyse von Laminaten (Improvement of the non-linear
failure analysis of laminates). In: Proceedings DGLR Deutscher
Luft- und Raumfahrtkongress, Friedrichshafen, 2005 [in
German].
[8] Knops M, Bo gle C. Gradual failure in bre/polymer laminates.
Compos Sci Technol 2006;66(5):61625.
[9] Schro der B. Untersuchungen zum Spannungs-Verformungs-Ver-
halten von unidirektional faserverstarkten Kunststoen bei Quer-
zugbelastung, bei Schubbelastung und bei kombinierter Belastung
aus Querzug und Schub. (Investigations on the stressstrain
behaviour of unidirectional bre reinforced plastics under trans-
verse tension, shear and under combined loadings of transverse
tension and shear). Diploma thesis, University Kassel, Konstruk-
tionstechnik, Fachgebiet Faserverbundtechnik (Prof. Puck); 1983
[In German].
[10] Swanson SR, Messick MJ, Tian Z. Failure of carbon/epoxy lamina
under combined stress. J Compos Mater 1987;21(7):61930.
[11] Camanho PP, Davila CG, Pinho ST, Iannucci L, Robinson P.
Prediction of in situ strengths and matrix cracking in composites
under transverse tension and in-plane shear. Composites A
2006;37(2):16576.
[12] Knops M. The Puck failure of failure in bre polymer laminates:
fundamentals, verication and applications. Berlin Heidelberg New
York: Springer-Verlag, 2007.
[13] Puck A. Festigkeitsanalyse von Faser-Matrix-Laminaten, Modelle fu r
die Praxis (Strength analysis of bre-matrix laminates, models for
practice). Mu nchen, Wien: Carl Hanser Verlag, 1996 (out of print) [in
German]. Available on-line as PDF le at www.klub.tu-darmstadt.de,
actual direct link to KLuB-downloads: http://www.klub.tu-darms-
tadt.de/forschung/download.php.
[14] Lutz G. The Puck theory of failure in laminates in the context of the
new guideline VDI2014 Part 3. In: Proceedings conference on damage
in composite materials, Stuttgart, 2006 (available online at
www.ndt.net, 11(12)).
Fig. A.1. Predicted fracture curve from Pucks IFF criterion.
Table A.1
Formulae of Pucks action plane related IFF criteria
IFF mode Stress exposure ratio f
E
Region of validity
A f
E

1
R
?k

R
?k
R
t
?
p
t
?k
_ _
2
r
2
2
s
2
21
_
p
t
?k
r
2
_ _
r
2
P0 (A.3)
B f
E

1
R
?k

s
2
21
p
c
?k
r
2
_ _
2
_
p
c
?k
r
2
_ _
r
2
< 0 and 0 6
r2
s21

6
R
A
??
js21cj
(A.4)
C f
E

s
2
21
4R
?k
p
c
?k
R
A
??

2

R
c
?

r2

r2
R
c
?

r
2
< 0 and 0 6
s21
r2

6
js21cj
R
A
??
(A.5)
With s
21c
R
?k

1 2p
c
?k
R
A
??
=R
?k
_
.
Table A.2
Recommended values for the inclination parameters
p
c
?k
p
t
?k
GFRP 0.25 0.30
CFRP 0.30 0.35
1964 A. Puck, M. Mannigel / Composites Science and Technology 67 (2007) 19551964

You might also like