Professional Documents
Culture Documents
art ic l e i nf o a b s t r a c t
Article history: A continuum damage model for simulating damage propagation of bonded joints is presented, introducing a
Accepted 1 February 2015 linear softening damage process for the adhesive agent. Material models simulating anisotropic non-linear
Available online 8 February 2015 elastic behavior and distributed damage accumulation were used for the composite adherends as well. The
Keywords: proposed modeling procedure was applied to a series of lap joints accounting for adhesion either by means
B. Composites of secondary bonding or co-bonding. Stress analysis was performed using plane strain elements of a
D. Cohesive zone model commercial finite element code allowing implementation of user defined constitutive equations. Numerical
E. Joggle lap joint results for the different overlap lengths under investigation were in good agreement with experimental data
Progressive damage in terms of joint strength and overall structural behavior.
& 2015 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ijadhadh.2015.02.001
0143-7496/& 2015 Elsevier Ltd. All rights reserved.
54 I.T. Masmanidis, T.P. Philippidis / International Journal of Adhesion & Adhesives 59 (2015) 53–61
As expected, a highly non-linear behavior under in-plane shear was the experimental stress–strain curve deviates from linearity close to
observed as well. coupon failure. Properties of the two resins are listed in Table 3.
To account for material non-linearity, incremental stress–strain
relations were implemented, retaining the validity of the general- 2.3. Progressive damage model
ized Hooke law for each individual interval as described in [3,4]:
E1t ν12 E2t 2.3.1. Composite adherends
dσ 1 ¼ dε1 þ dε2 Besides non-linear mechanical response, progressive damage
1 ðE2t =E1t Þν212 1 ðE2t =E1t Þν212
mechanics were also implemented in the FE modeling procedure.
ν12 E2t E2t
dσ 2 ¼ dε1 þ dε2 To account for the composite adherends progressive failure, the
1 ðE2t =E1t Þν212 1 ðE2t =E1t Þν212 Puck criterion [6] with the associated property degradation strategy
dσ 6 ¼ G12t dε6 ð1Þ is used. Details of the failure mode dependent stiffness degrada-
tion were described in [3] and are summarized for completeness
The tangential elastic moduli in the principal coordinate system of the
in Table 4. According to Puck theory, there are 5 ply damage modes,
orthotropic material, E1t (parallel to the fiber), E2t (transversely), G12t
two associated with either tensile or compressive fiber failure (FF)
(in-plane shear) were derived as follows by adopting the nonlinear
and three describing matrix cracking or inter-fiber failure (IFF);
constitutive model introduced by Richard and Blacklock [5]:
IFFA, -B, -C resulting mainly from a combination of transverse to the
Eoi εi fiber normal stress and in-plane shear.
σi ¼ h ni ið1=ni Þ ; i ¼ 1; 2; 6 ð2Þ
1 þ Eoi εi =σ oi Index (k) in the above relations refers to an arbitrary load step
after failure has been detected. The degradation factor, η r 1, multi-
By differentiating Eq. (2) one has: plying the engineering elastic constants to account for damage
ni ðð1=ni Þ þ 1Þ growth in the ply is given by [6]:
dσ σi
Eit ¼ i ¼ Eoi 1 i ¼ 1; 2; 1 ηr
dεi σ oi ηðk 1Þ ¼ þ ηr ð4Þ
1 þ cðf EðIFFÞ 1Þξ
ðk 1Þ
n6 ðð1=n6 Þ þ 1Þ
dσ 6 σ6
G12t ¼ ¼ Go12 1 ð3Þ
dε6 σ o6 where fE(IFF) is the failure effort as calculated by Puck's matrix failure
criterion while c¼5, ξ ¼3 and ηr ¼ 1 10 6 are the values of the
A summary of the numerical values for all constants in elasticity parameters of Eq. (4).
expressions can be found in Table 1; they were derived through
non-linear regression on the experimental data. Mean values for 2.3.2. Polymer resin
tensile and compressive strength properties in the fiber direction To account for the adhesive paste progressive damage (micro-
(XT, Xc), transversely to the fibers (YT, Yc) and in shear (S) for the cracking), since a brittle isotropic adhesive material is assumed, the
composite tested are given in Table 2. The relatively low elastic paraboloid failure surface criterion by Stassi D' Alia [7], adapted for
properties of the adherend are due to the wet hand-layup manu- generalized plane strain, is implemented:
facturing technique, characteristic of the in-situ patching procedure
of the industrial partner; typically this results in a fiber weight σ x σ y 2 þ σ y σ z 2 þ ðσ z σ x Þ2 þ 6τ2xy þ 2σ u ðR 1Þ σ x þ σ y þ σ z 2Rσ 2u r 0
fraction of ca. 51%. Mean values were deduced from 5 tests for each ð5Þ
specimen type while engineering elastic constants were derived as
where σ u represents the adhesive tensile strength and R, expressing
suggested by relevant standards.
the strength differential effect, is the ratio of compressive to tensile
failure stress. Here R is calculated in terms of the measured values
2.2. Polymer matrix and adhesive resin properties
Table 3
The epoxy resin, used as adhesive for the secondary bonded Engineering elastic constants and failure stresses for the polymer systems.
specimens, is HUNTSMAN XD 4734 with XD 4741-S hardener cured
E [GPa] G [GPa] σu [MPa] τu [MPa]
at 80 1C for 1 h. The response of both the, previously described,
polymer matrix of the adherends and adhesive resin was found to be Araldite LY3505/Aradur 3405 3.98 1.48 56.94 51.64
slightly non-linear especially under shear stressing. In this work the XD 4734/XD 4741-S 4.01 1.39 35.29 39.94
epoxy resins are assumed to have linear behavior until failure since
Table 1
Elasticity constants for the non-linear model, Eq. (3), of UD Glass/Epoxy composite. Table 4
Progressive stiffness degradation model for the
ν12 ¼ 0:26 composite adherends.
IFF(A) EðkÞ
2 ¼η
ðk 1Þ
E2
GðkÞ
12 ¼ η
ðk 1Þ
G12
Table 2
Failure stresses for the Composite material (in MPa). IFF(B) GðkÞ
12 ¼ η
ðk 1Þ
G12
IFF(C) EðkÞ
2 ¼ 10
10
E2
XT XC YT YC S
GðkÞ
12 ¼ 10 10
G12
558.60 411.12 40.00 128.14 38.42
I.T. Masmanidis, T.P. Philippidis / International Journal of Adhesion & Adhesives 59 (2015) 53–61 55
of shear strength τu and σ u of the adhesive. As it can be readily A continuum damage approach has been also introduced in [8],
derived from Eq. (5): where a stress/strain softening relationship is used as well. There
2 the area under the stress/strain curve was related to the critical
τu energy release rate by introducing a characteristic length in order
R¼3 ð6Þ
σu to transform displacements to strains.
When failure is detected a softening process is imposed to the
adhesive to account for damage evolution. For the kth step, after
failure has been detected, the degraded moduli are calculated by: 3. Coupon description and FE implementation
ðk 1Þ
EðkÞ ¼ 1 d E The lap joint geometry presented here, see Fig. 2, is usually
referred to as joggle lap joint (JLJ) and is often encountered in joints
GðkÞ ¼ EðkÞ =2ð1 þ νÞ ð7Þ where a smooth final surface is required, such as wind turbine rotor
blade restoration procedures. The presence of the joggle increases
The damage index, d, is calculated by considering the linear
the complexity of the problem due to the curvature; a study in
softening process shown in Fig. 1 for the equivalent stress and
which the JLJ configuration was also considered can be found in [9].
strain:
Total length and width of the coupons are equal to l¼500 mm and
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 w¼25 mm while the curvature length equals to a¼15 mm. In both
σ eq ¼ σ x σ y 2 þ σ y σ z 2 þ ðσ z σ x Þ2 þ 6τ2xy ð8Þ secondary bonded and co-bonded joint types the overlap length
2
(2c) varied from 50 to 200 mm in steps of 50mm, see Fig. 3.
An alternative approach concerning coupon geometry could be
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 3 to keep constant the ratio of overlap to total specimen length;
εeq ¼ pffiffiffi εx εy 2 þ εy εz 2 þ ðεz εx Þ2 þ γ 2xy ð9Þ another, to keep constant the distance of the joint edges from the
2ð1 þ νÞ 2
clamp area, both resulting in specimens of varying overall length.
As it can be readily proved from the above equations, σ eq ¼ Eεeq Nevertheless, a preliminary numerical investigation has revealed
and thus the scalar parameter d of Eq. (7) expressing the damage no significant differences in the ultimate load capacity and thus
accumulation for the current load step (k) is given by: results concerning joint behavior from the comparison between
the various overlap lengths presented in Section 4 are believed
ðkÞ
εequ εðkÞ
eq εeqo unaffected of said geometry variations.
d ¼ ðkÞ ð10Þ
εeq εequ εeqo All JLJ coupon adherends consist of 2 layers of the Glass/Epoxy
UD previously mentioned, fibers directed along the coupon long
where εðeqkÞ is the equivalent strain at that load step, εeqo ¼ εðeqnÞ is the axis, with a nominal total thickness equal to tc ¼1.744 mm.
equivalent strain of the material at the (nth) load step when the
failure criterion was satisfied and the softening process began and
3.1. Secondary bonded coupons
finally εequ is its maximum value.
Therefore, by updating the stiffness matrix, the stress for this
In the coupons manufactured with secondary bonding the
step is calculated by:
adhesive thickness was measured with the use of a digital caliper
σ ðkÞ ðkÞ ðkÞ
i ¼ C ij εj with i; j ¼ 1; …; 6 ð11Þ and by analyzing coupon free edge pictures with Mathworks
Matlab Image Processing toolbox, resulting in an average thickness
The slope of line AB in Fig. 1 defines the new stiffness of the of ta ¼1.30 mm for all coupons. For each overlap length, the curved
degraded material for a load step during the softening process. The adherend plate was manufactured by the industrial partner using
value σ eqo is of the equivalent stress when the failure criterion is a mold and was bonded to the already cured flat adherend plate
satisfied (nth load step) and therefore, it can have different values using spacers to ensure even distribution of adhesive thickness.
at the various elements depending on the current stress combina-
tion when failure is detected.
On the other hand, the value of the maximum equivalent
strain,εequ , for which complete failure occurs is assumed dependent
only on joint geometry and the adhesive material; it is derived by
adapting numerical simulations on the observed experimental
behavior of an arbitrary overlap length specimen from each joint
type. See Section 3.3 for details.
Fig. 1. Softening stress-strain curve after failure onset of the adhesive. Fig. 3. Adhesively bonded JLJ coupons of varying overlap length.
56 I.T. Masmanidis, T.P. Philippidis / International Journal of Adhesion & Adhesives 59 (2015) 53–61
Specimens were then cut using water jet. The adhesive paste spew, determined by:
present in all coupons, was included in the FE models, see Fig. 4,
since it was found to drastically influence the predicted joint t c Af tr
tr ¼ and t ec ¼ t c ð12Þ
behavior. 2 ρf 2
t r /2
tf
tc t ec
t r /2
Fig. 6. Composite adherend modeled as (a) two homogeneous UD plies (b) a set of resin and fiber layers and (c) as an equivalent composite layer and a resin layer at the
interface.
I.T. Masmanidis, T.P. Philippidis / International Journal of Adhesion & Adhesives 59 (2015) 53–61 57
Fig. 9. Comparison of numerical predictions and test results for the adhesively bonded JLJs.
Fig. 10. Location of strain gauges overlaid on the 2c¼ 150 mm overlap FE model.
Fig. 11. Measured vs. FEA calculated strains at the edge of lower adherend. Fig. 12. Measured vs. FEM calculated strains at the middle of the overlap.
thorough study of this mechanism should be undertaken in the slope change in the FEA curve indicates that debonding onset is
future. predicted at somewhat lower loads by the numerical model; to
The sudden load drop observed in the experimental curves at a further verify and validate its predictions, strain measurements were
magnitude around 4 kN marks the onset of debonding. The observed performed in some specimens of 150 mm overlap.
I.T. Masmanidis, T.P. Philippidis / International Journal of Adhesion & Adhesives 59 (2015) 53–61 59
Fig. 13. Comparison of numerically predicted and experimental damage patterns. (For interpretation of the references to color in this figure legend, the reader is referred to
the web version of this article.)
Fig. 14. Progressive debonding of the JLJ coupon at 20, 50, 75, 95 and 100% of the ultimate load.
Strain gauges 5 mm long were placed on the upper adherend in according to Puck criteria; FFT and FFC denote Tensile and
the middle of the overlap length, i.e. c¼75 mm, and on the lower Compressive Fiber Failure respectively while IFFA, IFFB and IFFC
adherend at a distance of s¼5 mm from the edge as shown in Fig. 10. refer to the 3 matrix failure modes as described in Section 2.3.1. It
Their measurements were compared with the strains calculated should be noted that even the trend for the adhesive spew to
from FEA at the corresponding nodes of the JLJ model. The strain remain attached to the upper adherend after complete debonding,
gauge placed at the lower adherend edge measures zero strain when see Fig. 8, was reproduced for all joints.
total debonding of the first 10 mm of the joint has occurred. A more detailed sequence of damage progression in the JLJ
As it is seen in Fig. 11, complete debonding of the overlap edge coupon of 150 mm overlap is also displayed in Fig. 14 showing
area was indeed predicted by the FE model. However the softening consistency with the observed fracture surfaces of the repair
procedure, eventually leading to the complete debonding, begins coupons indicating light fiber tear out, see Fig. 8.
at lower load levels than the experimentally observed abrupt
failure of the area. The strain gauge placed on the upper adherend 4.3. JLJ coupons joined by co-bonding
was strained up to 14 kN, see Fig. 12, indicating that the middle of
the overlap is close to the region to be debonded just before Comparison of experimental results in the form of load–
complete joint failure. This was verified by the FE model strain displacement curves from tests of JLJ coupons produced by co-
calculations and also by damage progression patterns, as those bonding and numerical predictions are presented in Fig. 15, where
shown in Fig. 13 where the debonding pattern is very similar to FEA curves compare well with experimental ones. The predictions
the experimental observation. The red colored elements in the FE of the Cohesive Zone Model (CZM), described in Section 3.5, are
model correspond to failed adhesive material while other colors also plotted along with the ones of the Continuum Damage Model
correspond to the various failure modes of the adherends (CDM), introduced in this work.
60 I.T. Masmanidis, T.P. Philippidis / International Journal of Adhesion & Adhesives 59 (2015) 53–61
Fig. 15. Numerical against test results for JLJs manufactured with co-bonding.
The trend to reach maximum strength restoration at 100 mm at the strength recovery that can be achieved using JLJ joint config-
was closely predicted as well by the FE results of both the proposed uration. In the current study experiments showed no increase in joint
model and the use of cohesive elements. It is of great significance to strength for overlap length greater than 150 mm or even 100 mm for
note that the time needed by the solver for the fully converged CZM the co-bonded coupons and this behavior was closely predicted by the
analysis approach, in some cases was even 7 times greater than the FE analysis.
one required for the proposed procedure.
Acknowledgments
5. Conclusions
Financial support by Compblades Ltd. (www.compblades.com)
A bilinear softening model combined with a failure criterion through contract D357-2011 with the Research Committee of the
suitable for brittle polymers has been introduced to account for University of Patras is gratefully acknowledged. The research work
damage progression and accumulation phenomena associated with was also partially funded by EYDE-ETAK of the Greek Ministry of
lap joints failure. Interaction of all stress components was taken into Development, in the frame of SYNERGASIA 2011 under contract
account using equivalent stress–strain response at the crack tip to ΣYN11_7_1000 (REWIND).
drive the softening procedure.
In addition to the proposed model, implementation of a progres- References
sive damage strategy for the composite adherends along with material
and geometrical nonlinearity resulted in load–displacement curves [1] Banea MD, Da Silva LFM. Adhesively bonded joints in composite materials: an
that showed good agreement with the experimental response. Esti- overview. Proc IMechE Part L: J Mater.: Des. Appl 2009:1–18.
mating for an arbitrary overlap length the only model parameter that [2] Xiaocong He. A review of finite element analysis of adhesively bonded joints.
Int J Adhes Adhes 2011;31:248–64.
is not related to the adhesive effective properties, resulted in satisfac- [3] Antoniou AE, Kensche C, Philippidis TP. Mechanical behavior of glass/epoxy
tory predictions for the other cases as well. Comparison of numerical tubes under combined static loading. Part II: Validation of FEA progressive
predictions from cohesive zone models and the current approach was damage model. Compos Sci Technol 2009;69:2248–55.
[4] Philippidis TP, Antoniou AE. A progressive damage FEA model for glass/epoxy
also favorable. Furthermore, the lack of need for a predefined crack
shell structures. J Compos Mater 2013;47(5):623–37.
path and use of generalized plane strain with a relatively coarse mesh [5] Richard RM, Blacklock JR. Finite element analysis of inelastic structures. AIAA
make the proposed FEA procedure adaptable for potentially modeling 1969;7:432–8.
large scale repair patches of complex geometries. [6] Puck A, Shürmann H. Failure analysis of FPF laminates by means of physically
based phenomenological models. Compos Sci Technol 2002;62:1633–62.
Numerical simulation provided satisfactory results for ultimate [7] Stassi F-D.' Alia. Teoria Della Plasticita e Sue Applicazioni. G. Denaro Ed.
loads in most cases. Test results demonstrate the existence of a limit Palermo 1958.
I.T. Masmanidis, T.P. Philippidis / International Journal of Adhesion & Adhesives 59 (2015) 53–61 61
[8] de Moura MFSF, Chousal JAG. Cohesive and continuum damage models applied to [13] Tsouvalis NG, Anyfantis KN. Determination of the fracture process zone under
fracture characterization of bonded joints. Int J Mech Sci 2006;48:493–503. Mode I fracture in glass fiber composites. J Compos Mater 2011;46(1):27–41.
[9] Taib AA, Boukhili R, Achiou S, Gordon S, Boukehili. H. Bonded joints with [14] Anyfantis KN, Tsouvalis NG. Experimental and numerical investigation
composite adherends. Part II. Finite element analysis of joggle lap joints. Int of Mode II fracture in fibrous reinforced composites. J Reinf Plast Compos
J Adhes Adhes 2006;26:237–48. 2011;30(6):473–87.
[10] Puppo AH, Evensen HA. Interlaminar shear in laminated composites under [15] Neto JABP, Campilho RDSG, da Silva LFM. Parametric study of adhesive joints
generalized plane stress. J Compos Mater 1970;4:204–20. with composites. Int J Adhes Adhes 2012;37:96–101.
[11] VDI 2014 Part 3. Development of FRP components Analysis. Verein Deutscer
Ingenieure e.V., 2006.
[12] Alfano G, Crisfeld MA. Finite element interface models for the delamination
analysis of laminated composites: mechanical and computational issues. Int
J Numer Methods Eng 2001;50:1701–36.