You are on page 1of 14

ARTICLE IN PRESS

International Journal of Adhesion & Adhesives 27 (2007) 505518 www.elsevier.com/locate/ijadhadh

Modelling the environmental degradation of adhesively bonded aluminium and composite joints using a CZM approach
C.D.M. Liljedahla, A.D. Crocombea,, M.A. Wahaba, I.A. Ashcroftb
b

School of Engineering, University of Surrey, Guildford, GU2 7XH, UK Wolfson School of Mechanical and Manufacturing Engineering, Loughborough University, Loughborough, LE11 3TU, UK Accepted 19 September 2006 Available online 17 January 2007

Abstract The long-term durability of adhesively bonded aluminium, composite and dissimilar substrate joints exposed to humid environments has been investigated. Failure of the joints was modelled with a cohesive zone model (CZM) approach where the governing parameters were determined from fracture mechanics test specimens saturated in a range of humid environments. The reduction in residual strength of an aluminium single lap joints (SLJ) immersed in de-ionised water was successfully predicted. For joints submerged in tap water the degradation was faster than predicted and there were signs of corrosion. XPS analysis carried out on the failure surfaces indicated that the more severe degradation might have been due to cathodic delamination. There was some discrepancy between the experimental and the predicted results of the aged composite SLJ. Composite failure might have contributed to this, which was not included in the modelling. Large residual stresses were induced in the dissimilar substrate joints due to the mismatch of coefcients of expansion of the substrates. The evolution of the total residual strains (thermal and swelling) was modelled. The residual stresses were seen to relax but were still signicant even after exposure for a year. The predicted degradation overestimated the residual strength of the double lap joints (DLJ). It is suggested that this may be due to a residual stress-enhanced degradation mechanism. r 2006 Elsevier Ltd. All rights reserved.
Keywords: Aluminium and alloy; Composite; Durability; Cohesive zone model; Progressive damage; Cathodic delamination; Residual stress

1. Introduction Major advantages of adhesive bonding to other forms of joining are an improved mechanical response and the ability to bond dissimilar materials. The major concern, which inhibits a more wide-spread use of this technology is their long-term durability in humid environments. A related issue is the effect of the residual stresses induced when bonding dissimilar materials. When a joint, bonded with a thermosetting adhesive, is allowed to cool from the curing temperature, residual thermal strains are induced. The thermal strains may reduce the durability of the joint. The thermal expansion of a material can be assessed by measuring the deection of a bi-material beam [1,2]. Peretz and Weitsman [3] have
Corresponding author. Tel.: +44 (0)1483689194; fax: +44 (0)1483306039. E-mail address: a.crocombe@surrey.ac.uk (A.D. Crocombe).

measured the coefcient of thermal expansion directly by means of strain gauges. The coefcient of thermal expansion of FM73 (the adhesive used in this work) was found to be 6.6 105 1C1. Upon moisture ingress the adhesive swells. The swelling may also cause signicant residual stresses in adhesively bonded joints. The moisture expansion of FM73-M has been determined from swelling of bulk adhesive sheets by Romanko and Knauss [4]. The coefcient of swelling was found to be 0.0022%1. The swelling of DGEBA/DDA epoxy resin during hygrothermal ageing has been investigated by Xiao and Shanahan [5]. The swelling coefcient was seen to be slightly dependent on the temperature. The average swelling coefcient was found to be 0.0016%1. Cabanelas et al. [6] have studied the water absorption in polyaminosiloxaneepoxy thermosetting polymers. The moisture uptake was determined with gravimetric measurements. The fractional volume change due to water uptake was determined through near-infrared spectroscopy. The

0143-7496/$ - see front matter r 2006 Elsevier Ltd. All rights reserved. doi:10.1016/j.ijadhadh.2006.09.015

ARTICLE IN PRESS
506 C.D.M. Liljedahl et al. / International Journal of Adhesion & Adhesives 27 (2007) 505518

equilibrium moisture content was around 2.5% and the coefcient of swelling was 0.0053%1. The residual stress may relax due to creep if stored after cure before testing. Thus, it is of importance to determine the viscoelasticplastic properties of the adhesive to be able to predict the long-term durability accurately. The effect of moisture on the viscoelastic shear properties of FM73 and FM300 M has been investigated by Jurf and Vinson [7]. The aim of the investigation was to determine the shear modulus and the time-dependent shear creep compliance for a wide range of temperatures. This was done for both adhesives under dry conditions and after exposure to 63%RH and 95%RH to 70% saturation. The exposure time was calculated from the Fickian diffusion coefcient derived from gravimetric experiments. It was found that the effect of moisture had the same effect on the mechanical properties as an increase in temperature. Experimental data on butt joints obtained by Reedy and Guess [8] suggested that residual stress generated by cooling after cure had little effect on their joint strengths. Stress relaxation data for the adhesive used showed that after loading the adhesive to yield, the stress level decreased by 30% over a period of 30 min. This showed that stress relaxation might be the reason why residual stresses often have a much smaller effect on the joint strength than expected. Transient hygrothermal stresses in bre-reinforced composites have been studied by Vaddadi et al. [9]. However, they did not include any moisture dependent or viscoelastic material behaviour. Liljedahl et al. [10] have carried out a fully coupled diffusion stress analysis of the hygrothermal stresses in bonded steel joints. It was found that the residual stresses relaxed signicantly and thus did not decrease the residual strength signicantly. The degrading effect of water on structural adhesive joints has been investigated by Gledhill and Kinloch [11]. The initial failure stress of a mild steel-epoxy butt joint was 37.5 MPa. For the joints exposed to immersion in distilled water, the measured strength decreased considerably; the higher the immersion temperature, the more rapid was the decay. The initial locus of failure was cohesive through the adhesive. The locus of failure for joints immersed in water was interfacial at the edges but cohesive in the middle. Longer exposure time gave a larger interfacial zone. After saturation no further degradation occurred. Gedhill et al. [12] suggested that the durability can be predicted by combining water diffusion data with fracture mechanics. The diffusion in the bulk adhesive was determined using gravimetric experiments. The moisture uptake data were tted to the Fickian diffusion relation. Applying a fracture mechanics approach, it was found that the fracture stress could be predicted if the water concentration in the adhesive was known. Crocombe [13] has developed a framework to assess the effect of environmental degradation in adhesively bonded joints and subsequent cohesive failure. The moisture ingress in the joint was assumed to be Fickian. Transient nite element analysis was used to determine the moisture

concentration in the joint due to diffusion. Modelling of lap joints bonded with FM1000 was then undertaken. The author suggested that the joint failure occured at the centre of the exposed joint, where the adhesive was dry and less ductile. The joint strengths were predicted to within a few percent. A cohesive zone model (CZM) modelling methodology has been shown to be a versatile approach to predict the durability of adhesively bonded joints exposed to humid environments [1416]. The aim of the current study was to investigate use of a CZM approach to predict aluminium and composite single lap joints (SLJ) and aluminium composite double lap joints (DLJ). This extends the previous work of the authors, which included modelling degradation that was related directly to the moisture concentration with another adhesive system [10] and the modelling of aluminium SLJ and L joints using the same adhesive as this study [16]. In this latter work, approaches for including corrosion and stress-accelerated aging were introduced. In this current work, as well as introducing other joint congurations, evidence of this corrosion failure mechanism was detected with X-ray photoelectron spectroscopy (XPS). Further, the effect of the large residual strains induced in dissimilar substrate joints on the damage initiation and crack propagation was investigated. 2. Experimental methods and results In this study the effect of humid environments on the durability of adhesively bonded aluminium and composite joints have been investigated. The environments considered were 80%RH/70 1C, 96%RH/50 1C and immersion in deionised and tap water, both at 50 1C. 2.1. Material properties of the joint constituents During this study mixed mode exure (MMF) adhesive joint fracture specimens, notched coating adhesive (NCA) specimens, SLJ (both cfrpcfrp and aluminiumaluminium) and dissimilar substrate (aluminiumcfrpaluminium) DLJ were tested. A heat cured lm adhesive (Cytec FM73) was used to bond all specimens. 2.1.1. Adhesive The moisture ingress in the adhesive was determined by gravimetric experiments on bulk adhesive samples. The experimental details are given elsewhere [16] and the data are summarised in Table 1. The coefcient of thermal expansion (CTE) and the stress-free temperature (100 1C) were determined by measurement of a bi-material beam and are given in Table 2. Further details are can be found elsewhere [17]. Hygroscopic expansion (swelling) of the adhesive was assumed to be isotropic and was therefore deduced by measuring the expansion of a bulk adhesive dumbell specimen (used for subsequent mechanical testing) with a

ARTICLE IN PRESS
C.D.M. Liljedahl et al. / International Journal of Adhesion & Adhesives 27 (2007) 505518 Table 1 Fickian diffusion parameters for the adhesive Adhesive FM73 FM73 FM73 Relative humidity (%) Immersion 95.8 79.5 Temperature (1C) 50 50 70 Saturation content (%) 3.5 2.2 1.2 Diffusion coefcent (1014 m2/s) 52.2 50.2 790.0 507

Table 2 Coefcient of thermal expansion and hygroscopic swelling expansion for the adhesive Coefcient of thermal expansion (1C1) 7.8 105 Coefcient of hygroscopic expansion (%1) 0.0021

Table 3 Mechanical properties for the adhesive at various moisture concentrations at RT Material Environment Moisture content (%) 0.0 1.2 2.2 E (MPa) suts (MPa)

FM73 FM73 FM73

Dry 80% RH/70 1C 96% RH/50 1C

2000 1700 1500

45 42 38

0.5
Table 4 Creep parameters at 50 1C (force (N), length (mm), time (s), stress (MPa)) Condition A 6.5458 1011 1.3493 1010 1.5848 1010 N 4.75 4.75 4.75 m 0.6 0.5 0.5

0.4

Strain (%)

0.3

Dry 96% RH Immersion

0.2 specimen 1 specimen 2 specimen 3 specimen 4 least square 0 0.5 1 1.5 Moisture content (%) 2 2.5

0.1

creep tests on bulk adhesive specimens. The details are given elsewhere [16]. The data were tted to the following power law and the best-t parameters for the different environments can be seen in Table 4. _ creep Aqn tm , where a dot represents differentiation with respect to time, q is equivalent stress, t the time and A, n and m are constants. 2.1.2. Composite The carbon bre reinforced polymer (cfrp) composite was received as 0.125 mm thick prepeg (IM7/8552) from the manufacturer, laid up unidirectionally and cured in an autoclave. The cure cycle was 275 min and the highest temperature was 177 1C as suggested by the manufacturer (Hexel). The laminates were then cut to size from the cured lms. The diffusion properties transverse and parallel to the bre direction were determined with gravimetric experiments and are given in Table 5. The CTE for the composite used in the analyses, were 0.06 and 2.86 105 K1 parallel and transverse to the bres, respectively [18]. The CHE of the composite was determined both in the transverse and parallel direction to the bre as given in Table 6. The expansion coefcients in the parallel direction were much smaller than in the transverse direction as expected as the bres (which govern the behaviour in that direction) have a

Fig. 1. Expansion due to moisture ingress in bulk adhesive samples.

shadowgraph (Mitutoyo Prole Projector PJ 300). The samples were removed periodically from the ageing environment, and the length of the specimens measured. The 95.8%RH/50 1C environment was used as the diffusion process was slower and the equilibrium moisture content higher than in the 80%RH/70 1C environment. The swelling due to moisture ingress in the adhesive can be seen in Fig. 1. The swelling was seen to be proportional to the moisture content. The coefcient of hygroscopic expansion (CHE) derived from the data is given in Table 2. The effect of moisture on the bulk adhesive properties was determined by undertaking tensile tests of the bulk adhesive dumbbell specimens aged in the different environments. The experimental details are given elsewhere [16] and the data obtained are summarised in Table 3. From the table it can be seen that both the modulus and the ultimate tensile strength reduced in the presence of moisture. The creep properties of the adhesive after saturation in the aging environments were determined by undertaking

ARTICLE IN PRESS
508 C.D.M. Liljedahl et al. / International Journal of Adhesion & Adhesives 27 (2007) 505518 Table 5 Fickian diffusion parameters for the composite Material Relative humidity (%) Temperature (1C) Saturation content (%) Diffusion coefcent (1014 m2/s)

Fickian diffusion parameters for the composite IM7/8552 95.8 IM7/8552 95.8 IM7/8552 79.5 IM7/8552 79.5

50 50 70 70

1 1 0.8 0.8

20 (transverse) 70 (parallel) 50 (transverse) 140 (parallel)

Table 6 Hygroscopic expansion of the composite Parallel CHE (%-1) 0.0 Transverse CHE (%1) 0.0046

Table 7 Mechanical properties for the composite E11 (GPa) 168 E22, E33 (GPa) 9.8 G12, G13 (GPa) 4.8 G23 (GPa) 3.2 n12, n13 0.31 n23 0.52

very low coefcient of thermal expansion and do not absorb a lot of moisture. In the transverse direction, the properties of the composite are governed by the epoxy matrix, which has a higher CTE and absorbs much more moisture than the carbon bres. Most of the mechanical properties (E11, E22, E33, G12, G13) for the unidirectional composite (IM7/8552) were determined for dry condition and for saturation at various relative humidities. The mechanical properties for the composite did not change signicantly in the presence of moisture for the temperature range considered (from 20 to 70 1C). This limited change was probably because the composite only absorbed a small amount of moisture and was cured at a high temperature. The dry RT values were thus used in the modelling discussed in Section 3 in this study and are given in Table 7. The remainder of the mechanical properties used were as found by Schon et al. [19]. 2.1.3. Aluminium The elastic modulus of the aluminium was determined with a tensile test (BS EN 1386:1997) and was found to be 68400 MPa. The Poissons ratio used was 0.33. The CTE used for the aluminium alloy was 2.36 105 K1. 2.2. Joint tests First, fracture mechanics tests were undertaken to characterise the fracture toughness after aging in the different humidity environments. A variety of lap shear joints were also tested. Both aluminium and composite SLJ and dissimilar substrate (aluminiumcompositealuminium) DLJ were aged and tested. Generally at least three replicates were used to assess residual strength. A few of the exploratory aluminium SLJ tests had less. The dimensions of the lap shear tests are shown in Fig. 2. The adhesive thickness was 0.15, 0.06 and 0.13 mm for the aluminium SLJ, the composite SLJ and the DLJ, respec-

tively. The thinner adhesive layer in the composite SLJs was not planned but arose from difculties in the manufacturing procedure Before bonding, the aluminium alloy substrates were pretreated in a chromic acid etch solution (CAE) according to DEF STAN 03-2/issue 3 and the composite substrates were degreased with acetone. All joints were cured for 1 h at 120 1C. 2.2.1. Fracture energy tests Fracture energy tests were carried out to determine the moisture-dependent cohesive zone model parameters in the modelling section. A MMF specimen was used for unaged and saturation in the 80%RH and 96%RH environments. For immersion in de-ionised water a NCA specimen was used. The experimental details and results can be found elsewhere [10,16]. In brief, these joints are aged as openfaced specimens (a layer of cured adhesive on a single substrate). The adhesive layer of the NCA specimen is then scored (notched) and the specimen loaded in simple tension until the adhesive layer debonds. For the MMF specimen another substrate is secondary bonded to the adhesive layer forming a laminate. This laminate is loaded in bending in the usual way until debonding occurs. In both cases the debond load can be used to determine the environmentally degraded interfacial strength. 2.2.2. Aluminium SLJ The aluminium SLJs were manufactured individually and not cut from large-bonded plates as this caused bending. The joints were immersed in water at 50 1C. The failure surfaces were rst analysed by visual inspection. The amount of apparent interfacial failure in relation to the failure load can be seen in Fig. 3. In the gure it can be seen that the there was a relation between the amount of cohesive failure and the failure load. After aging for 2 weeks, a sharp decrease in failure load was seen, which, after another 2 weeks, was apparently

ARTICLE IN PRESS
C.D.M. Liljedahl et al. / International Journal of Adhesion & Adhesives 27 (2007) 505518 509

38

0.15 0.01 3.16

a
38

b
Compression side 2.01 12.5

2.01.

0.06 0.01

c
84 100

Tension side 3.97

0.13 0.04

Fig. 2. Dimensions of the lap joints [width 25 mm] (all dimensions in mm) (a) AlAl SLJ, (b) crfpcfrp SLJ and (c) AlcrfpAl DLJ.

0 12 10 Failure load (kN) 8 6

50

100

150

200

250

300

350 100 12 10 Cohesive failure (%) Failure load (kN) 8 6 4 2


Predicted, 96% RH Predicted, de-ionised Experimental, 96%RH Experimental, de-ionised Experimental, original

Immersed Dry

80

60

40 4 2 0 0 50 100 150 200 250 Time (days) 300 350 20

0 0 0 10 20 30 Immersion time (week) 40 50

Fig. 3. Failure load and apparent cohesive failure for the aluminium SLJ.

Fig. 4. Repeat aluminium SLTs in de-ionised water and 96%RH are compared with the original results and predictions.

recovered. Both the results for the failure load and the amount of cohesive failure followed the same trend. Storing the joints at 50 1C in dry conditions for prolonged periods did not affect the joint strength, i.e. there was no thermal degradation (Fig. 3). Additional tests, aging the samples in de-ionised and tap water and at 96%RH were undertaken for 2 and 12 weeks to investigate this behaviour further. These data, together with the original data are shown on Fig. 4. This gure also contains residual strength predictions that will be discussed later. Images of failure surfaces from the original tests and the repeats are compared in Fig. 5. It can be seen that the aluminium substrates from the original tests aged for a shorter time were much more corroded than the ones from the repeated test where de-ionised water was used. It can also be seen that the substrates from the original tests, aged for longer times, look similar to the ones tested in deionised water. In the original tests, undertaken by our research partners, separate baths were used for the

different ageing times and it is conjectured that tap water may have been used in the short exposure tests while deionised water may have been used for the longer exposure tests. The degradation of bonded joints exposed in tap water compared with distilled water has been shown to be more critical [20]. Specimens immersed in tap water failed early in the programme and specimens immersed in distilled water did not show any signicant reduction in strength after 7 years. Salts dissolved in tap water result in a higher conductance, which enhances corrosion. The conductance of de-ionised water and tap water was measured with a conductivity meter (Jenway 4310) and are given in Table 8. The conductance was several orders of magnitude higher in the tap water. Diffusion of cations in a saline solution has been shown to determine the rate of delamination [21]. The faster degradation might hence be due to an electrochemical process in the form of cathodic delamination as

ARTICLE IN PRESS
510 C.D.M. Liljedahl et al. / International Journal of Adhesion & Adhesives 27 (2007) 505518

4.5 4 3.5 3 Counts 2.5 2 1.5 1 0.5 0

x 104
O1S

Distance from free edge

C1s

Mg1s

Na1s Zn2p3

Ca2p N1s Al2p

1200

1000

800 600 400 Binding evergy (eV)

200

Fig. 6. Survey spectra for the XPS analysis on joints aged in tap water for 2 weeks.

Fig. 5. Aluminium SLJ aged in water, comparison between original tests and repeats.

Table 8 Conductance in de-ionised and tap water Water De-ionised Tap Conductance (mS) 0.83 496

suggested by Davis and Watts [22]. Cathodic delamination consists of the following reactions: electrons of the free exposed aluminium surface are dissolved Al ! Al3 3e and, in the relatively anodic region covered with epoxy, oxygen and water react with the electrons, and hydroxyl ions are produced. O2 2H2 O 4e ! 4OH .

Hydroxyl ions are deleterious [23] to the interface strength and the adhesive debonds and the newly exposed aluminium now acts as the cathode and the still-bonded overlap as the anode. This process continues until the whole overlap has debonded. To further investigate this, XPS was undertaken on samples aged for 2 weeks in tap water and de-ionised water. The XPS analyses were undertaken using a VG scientic ESCALAB MKII system operated in constant analyser mode. AlKa radiation was used in all analyses. The pass energies were set to 100 eV for the survey spectra and 20 eV for the high-resolution spectra. The fracture surface where apparent interfacial failure had occurred was line scanned from the exposed edge towards the centre of the overlap (Fig. 5). A spot size of 500 mm was used. The rst point was close to the edge and the next seven points moved progressively towards the centre of the joint. The survey spectra from the XPS analyses are shown in Fig. 6. The results for calcium and nitrogen are summarised in Fig. 7. Less aluminium and more carbon were detected as the scan moved further away from the exposed edge. This trend was expected as there was less water in the centre of the joint. The carbon may come from the adhesive or as contamination. Calcium may be used to indicate corrosion, as the negative hydroxyl ions produced in the corrosion process attract cations [24]. However, the calcium prole followed the same pattern as the moisture distribution and hence no denite conclusion can be drawn from the calcium. The higher amount of calcium seen for the sample immersed in tap water may hence be driven by water diffusion and not due to a charge carrying potential. The nitrogen is a label for the curing agent [25] and hence the adhesive. No nitrogen was seen close to the edge for the joint exposed to the tap water environment. This indicates a more interfacial failure for this joint, which might be due to corrosion.

ARTICLE IN PRESS
C.D.M. Liljedahl et al. / International Journal of Adhesion & Adhesives 27 (2007) 505518 511

3 2 %at 1 0 0 0.2 0.15 %at 0.1 0.05 0 0 1 2 3 4 5 Distance along the overlap (mm) 6 7
N1s-Tap N1s-DeI Norm Conc Ca2p-Tap Ca2p-DeI Norm Conc

Fig. 7. Summary of the XPS results for calcium and nitrogen.

12 10 Failure load (kN) 8 6 4 2 0 0 10 20 30 40 50 Time (week) 60 70 80

SLJ Comp 80% SLJ Comp 96% Pred 96% Pred 80%

Fig. 8. Test results for the composite SLJ and predictions.

Fig. 9. Failure surfaces of the composite SLJ joints.

2.2.3. Composite SLJ The composite SLJs were exposed to 80%RH at 70 1C and 96%RH at 50 1C. The results from the tests can be seen graphically in Fig. 8. The sharp drop after eight weeks for the joints exposed to 80%RH may indicate that the failure was governed by composite delamination. The failure surfaces of the tested joints can be seen in Fig. 9. It can be seen that there was some composite failure for some of the joints but in none of the joints was the region of composite failure larger than the region of adhesive failure. The delamination was more evident on the joints exposed at a higher relative humidity. This might suggest that they were caused by environmental degradation rather than manufacturing defects. 2.2.4. DLJ The DLJs were exposed to 80%RH at 70 1C and 96%RH at 50 1C. The results from the tests can be seen

in Fig. 10. The DLJ was the most complex conguration as it had both composite and aluminium substrates. This induced large residual stresses and also a variety of possible failure mechanisms. As seen in the Fig. 10, the failure load was very similar in the two environments. This indicates that the failure in these joints may not solely relate to the level of moisture. There are two other possible mechanisms, which are composite failure and stress-enhanced degradation due to the residual stresses. There was some composite failure on the failure surfaces as seen in Fig. 11. The highest amount of composite failure was seen on the dry samples and the samples aged in the 96%RH environment for 1 week. However, there were signs of composite failure on many of the other samples too. The strength for the dry DLJ aged at 70 1C increased (Fig. 10). This can probably be attributed to relaxation of the residual stresses and not to a post-cure effect as the

ARTICLE IN PRESS
512 C.D.M. Liljedahl et al. / International Journal of Adhesion & Adhesives 27 (2007) 505518

strength of the SLJs stored dry at an elevated temperature was nearly unaffected (Fig. 3).

3. Modelling The failure response of the joints was simulated using a CZM. A two-parameter CZM was used where separations in mode I, II and III were accounted for. The fracture energy in all modes was the same. This can be justied as
0 25 90 Failure load (kN) 20 15 10 5 0 0 50 100 150 200 250 Time (day) 300 350
80%RH 96%RH Dry

50

100

150

200

250

300

350 100

plasticity, which gives rise to mode-dependent fracture energy is included explicitly in the adhesive material model. The fracture energy in the CZM thus relates more closely to an intrinsic fracture energy. The CZM parameters are the energy (G0) and the tripping traction (su). The stiffness before unloading was set high to avoid any signicant compliance of the CZM element before the onset of damage (Fig. 12). The commercial FE (nite element) package, ABAQUS (Hibitt, Karlsson & Sorensen, Inc) was used for all the numerical modelling work. Non-linear springs (ABAQUS 6.5: 18.1.1) were used to specify the traction-separation law at the interface in the modeling.

Cohesive failure (%)

(MPa) u

80 70 60 50 40

o (kJm-2)

(mm) r
Fig. 12. Shape of the CZM used throughout the study.

Fig. 10. Failure load and amount of adhesive (cohesive) failure vs. aging time.

Fig. 11. Failure surfaces of the DLJ in 96%RH for 1 (a) and for 52 weeks (b).

ARTICLE IN PRESS
C.D.M. Liljedahl et al. / International Journal of Adhesion & Adhesives 27 (2007) 505518 513

CZM is CPU intensive and it was therefore convenient to model the joints in 2D if appropriate. Extensive diffusion and stress analysis without CZM elements were thus carried out to using a range of 2D and 3D elements, with and without residual strains to determine if 2D models could be used. This is discussed further in relation to each joint conguration. Firstly, the CZM parameters were determined for various moisture concentrations using the MMF and NCA specimens. These parameters were then used to predict the response of the SLJ. Finally, the response of the DLJ was predicted. The adhesive continuum was modelled using a linear DruckerPrager model with a friction angle of 32.51 [17] and moisture-dependent hardening, determined experimentally from the stressstrain test of bulk adhesive specimens. 3.1. Determination of moisture dependent CZM parameters The CZM parameters were determined from the MMF specimen for unaged conditions and for saturation at 80%RH and at 96%RH. The parameters for immersion in de-ionised water were determined from the NCA test. The tripping traction was determined by the onset of non-linearity in the load displacement response. The fracture energy (G0) was then determined by correlating the simulated loadcrack length characteristics with the experimental results. This gave a unique set of CZM parameters for each of the aging environments. These CZM properties operated in a region that ensured mesh independence. A more detailed discussion of the determination of the CZM parameters is given elsewhere [16]. The moisture-dependent fracture energy and tripping traction parameters determined from the MMF and NCA calibrations are shown in Fig. 13. The fracture energy is seen to decrease rapidly initially and then atten out whereas the opposite trend applies to the tripping traction. It can be seen that the data determined from the NCA and MMF tests appear to be entirely consistent. This is a feature that has been identied in related work using a different adhesive system. 3.2. Aluminium SLJ The CZM parameters determined from the fracture tests were used in this section to predict the durability of the SLJ. The prediction was undertaken in two steps. First, the moisture prole throughout the joint was predicted in a diffusion analysis using the determined diffusion parameters (Table 1). Then, the predicted moisture prole was input as a predened eld in the subsequent CZM of the SLJ as both the adhesive continuum and CZM parameters were a function of the moisture concentration (Table 3 and Fig. 13). It was found that the diffusion in the SLJ could be modelled accurately in 2D as the width was twice as long as the overlap and thus diffusion from the sides was less signicant. It also was found that neither thermal nor

0.5 3000 2500 Fracture energy (Jm-2) 2000 1500 1000 500 0 0.5

1.5

2.5

3 45 40 35 30 25 20 15 Tripping traction (MPa)

1.5 2 Moisture (%)

2.5

Fig. 13. Determined moisture-dependent tripping traction and fracture energy.

Fig. 14. The mesh used for the aluminium SLJ.

swelling residual strains were signicant and thus did not have to be included. The SLJ could be modelled accurately assuming plane strain conditions as the adhesive was constrained by the substrates and the adhesive layer was thin The mesh used for the SLJ throughout this section is shown in Fig. 14. The llet was modelled to get an accurate correlation between the model geometry and the experimental test specimen. The predictions of the residual strength of the aluminium SLJs aged in the 96%RH and by immersion in water are shown in Fig. 4. The prediction of joints submerged in deionised water and aged in the 96%RH were very good. The residual strength of the joints that were heavily corroded (short exposure times for the original tests) where however overestimated. The cathodic delamination degradation mechanism can be included in the modelling. This is however not done here. A model to include this behaviour has been developed and applied to an L joint submerged in tap water. The details can be found elsewhere [16].

ARTICLE IN PRESS
514 C.D.M. Liljedahl et al. / International Journal of Adhesion & Adhesives 27 (2007) 505518

3.3. Composite SLJ The approach to model the composite SLJ was the same as for the aluminium SLJ and the same 2D assumptions could be used. The residual stresses were also seen to be small. For the diffusion, the ingress of moisture through the composite was included and the different saturation levels in the composite and the adhesive were accounted for by using a mass diffusion analysis (ABQUS Users Manual 6.5: 6.8.1). A similar mesh to that used for the aluminium SLJ (Fig. 14) was used for this joint The prediction was expected to be accurate up to moderate moisture contents as the failure mode was mainly cohesive up to saturation in the 96%RH for the MMF specimens. For high-moisture concentrations the failure mode was apparent interfacial between the epoxy and the aluminium substrate for the MMF and hence the modelling of the cfrp SLJ would be expected to under predict the strength for high-moisture concentrations. The results can be seen in Fig. 8. It can be seen that the prediction of the dry strength was very good and the predictions for the 96%RH were as expected. The experimental data of the joints aged in the 80%RH environment were seen to degrade slowly initially and then at about 12 weeks the strength dropped drastically. This may be due to composite failure. However, this was not included in the model. This mechanism can however be included by dening the delamination strength for the composite with a mode dependent CZM. However, the experiments necessary to derive the mode-dependent CZM parameters were not within the remit of this work. This will form an aspect of future work. Also, the strength of the cfrp SLJ aged for 78 weeks increased. The mechanism for this increase is not understood. The data might not be representative or there might be a mechanism that increased the joint strength for long aging times that is not included in the model. 3.4. Aluminiumcompositealuminium DLJ The residual stresses in the DLJ are large due to mismatch in coefcients of expansions between the substrates. Results from both 2D and 3D analysis are reported for these joints. In the rst part of this section stress analysis results for the evolution of the residual stresses are given. In the second part progressive damage and failure prediction modelling of the DLJ is described. Linear material properties have been adopted for the stress analysis (3.4.1) but, as elsewhere in this paper, a non-linear DruckerPrager model has been used for the adhesive in the progressive damage modelling (3.4.2) 3.4.1. Stress analysis The mesh used for the 3D model used can be seen in Fig. 15. The smallest elements were 0.065 0.065 and 0.065 0.065 0.25 mm for the 2D and the 3D models, respectively. An analysis with a ner mesh (smallest

Fig. 15. Mesh used for the 3D model of the DLJ.

100 90 80 von Mises (MPa) 70 60 50 40 30 20 0 5 10 Distance along the overlap (mm) 15


centre 3D edge 3D plane strain plane stress plane stress-strain

Fig. 16. Comparison between 3D and 2D of stresses due to mechanical load (DLJ).

elements 0.0325 0.0325 mm) was also undertaken. This indicated that the coarser mesh was sufciently accurate. Eight-noded reduced integration elements (C3D8R) and four-noded rst order elements were used for the 3D and 2D models, respectively. Stresses on the mid plane of the adhesive, from various models at the unaged mechanical failure load are plotted in Fig. 16. The residual (thermal and swelling) strains have not yet been included. Both the peel and the shear stress were larger at the free edge due to the mismatch in Poissons ratios of the substrates (the failure was seen to initiate at this edge when modelling the failure with a CZM, Section 3.4.2). Three different 2D models were considered: plane stress in the whole joint, plane strain in the whole joint and plane stress in the substrate and plane strain in the adhesive (plane stressstrain). Both models with a plane strain assumption for the adhesive were seen to give a good correlation with the 3D model at both the compression and tension side (see Fig. 2 for the location of compression and tension sides) at the centre of the joint. The plane stress model overestimated the stress at the tension side. Higher levels of stress are seen on the

ARTICLE IN PRESS
C.D.M. Liljedahl et al. / International Journal of Adhesion & Adhesives 27 (2007) 505518 515

compression side as the stiffness in the loading direction is lower for the aluminium than for the composite (failure was seen to initiate on this side when neglecting residual strain and modelling the failure with a CZM model as shown in Fig. 22). 3.4.1.1. Evolution of residual stress in the DLJ. The stresses induced when the specimen was cooled from the curing temperature can be seen in Fig. 17 (creep was not included at this stage). The stresses are much more symmetric than those caused by the mechanical load (Fig. 16). The stress state was very similar at the edge and in the centre of the joint as the coefcient of thermal expansion for the both substrates was very similar in the width direction (Section 2.1). The plane strain analysis overestimated the adhesive stresses as neither the substrates nor adhesive could contract in the out-of-plane direction, as happened in practice. Assuming a state of plane stress however, resulted in the adhesive stresses being underestimated in the middle of the joint as the out-of-plane normal stresses in the adhesive are not modelled. The plane stressstrain model gave the best approximation as the substrates are not constrained in the width direction but the adhesive was constrained by the substrates. The residual stress was then determined after 84 days of aging in the 96%RH/50 1C environment, as seen in Fig. 18. The stresses through the centre of the joint were dominated by the mismatch of coefcient of thermal expansion in the overlap direction, but were reduced somewhat at the overlap ends due to swelling in the adhesive caused by the absorbed moisture in this region. However, the stresses at the edge increased as the composite experienced hygroscopic swelling in the width direction, which caused higher strains in that region, as the aluminium substrates did not swell. The peak stresses in the centre of the joint could be modelled fairly accurately by plane stress as was
von Mises (MPa)

45 40 35 30 25 20 15 10 5 0 0 5 Distance (mm)
Fig. 18. Comparison between 3D and 2D stresses due to residual strain after 84 days in the 96% RH (thermal and swelling strains).
edge 3D centre 3D plane stress plane strain plane stress-strain

10

15

50 45 40 von Mises (MPa) 35 30 25 20 15 10 5 0 5 10 Distance along the path (mm) 15


2 weeks 2 weeks (creep) 12 weeks 12 weeks (creep) 52 weeks 52 weeks (creep)

60
centre 3D edge 3D plane strain plane stress plane stress-strain

Fig. 19. Evolution of residual stresses (creep, plane strain).

50

von Mises (MPa)

40

30

20

10

0 0 5 10 Distance along the overlap (mm) 15

Fig. 17. Comparison between 3D and 2D of stresses due to thermal load (DLJ).

the case for the thermal strains alone (Fig. 17). However, unlike the thermal strains alone, the peak stresses at the overlap edges were closer to the plane strain and plane stressstrain solutions as the transverse swelling in the composite increased the out-of-plane stresses. A coupled diffusionstress analysis was carried out in 2D, where creep was included during aging at the elevated temperature. The analysis technique is described in detail in Liljedahl et al. [16] where it was applied to stressed SLJs. This was done to investigate if the residual stress relaxed signicantly during aging. The creep properties were as derived experimentally (Table 4). The residual stresses were seen to reduce to about half of the magnitude obtained when creep was not included, see Fig. 19. Note that the stresses in the centre of the joint reduce from 12 to 52 weeks. This is because moisture reaches this central region

ARTICLE IN PRESS
516 C.D.M. Liljedahl et al. / International Journal of Adhesion & Adhesives 27 (2007) 505518

at longer exposure times and the associated hygroscopic swelling counters the differential thermal strains. Overall, plane strain is seen to represent the stresses due to mechanical loading fairly accurately (Fig. 16). It can be concluded that residual stress has to be included in the CZM. Even when creep is included the residual stresses are signicant in comparison with the stresses due to mechanical loading. The thermal residual strains were overestimated when plane strain was used (Fig. 17) but the peak stresses from the plane strain model for the aged specimens correlate quite well with the 3D model (Fig. 18). Thus, the best 2D assumption for the DLJ is plane strain. 3.4.2. Progressive damage and failure prediction of the DLJ Plane strain was used in the 2D model as this was shown to provide the best representation of stresses in the previous section. The smallest elements in the 2D and 3D model were 0.025 0.025 mm and 0.10 0.20 0.40 mm (C3D8R), respectively. From the 2D (CPE4R) modelling it was deduced that the process zone was long. Larger elements could thus be used in the 3D model without violating the criteria for mesh independence (i.e. the process zone should be longer than a few elements). Modelling was carried out both with and without residual strains as these were seen to be signicant. The results from the 2D modelling can be seen in Fig. 20. The damage initiation (tripping: maximum cohesive stress is reached and the strength starts to deteriorate, Fig. 12) and crack propagation (release: when the cohesive strength has decreased to zero, Fig. 12) can be seen in Figs. 21 and 22 for a dry joint and a joint aged for 1 year at 95%RH/50 1C. For the dry joint it can be seen (Fig. 21) that the damage initiation started at the compression side when the residual strains were neglected and started on the tension side when the residual stains were included. The damage started at an earlier stage in the model where residual strains have been

0.35 0.3 Applied displacement (mm) 0.25 0.2 0.15 0.1 0.05 0 0 2 4 6 8 10 12 Distance along the crack path (mm) 14
Tripping Release Res Tripping Res Release

Fig. 21. The effect of the thermal strain on the crack propagation for a dry DLJ (plane strain).

0.2

Applied displacement (mm)

0.15

0.1

0.05

Tripping Release Res Tripping Res Release

0 0 2 4 6 8 10 12 Distance along the crack path (mm) 14

25

Fig. 22. The effect of the residual (hygroscopic and thermal) strains on the crack propagation for a DLJ aged for 365 days in the 96% environment (plane strain).

20

15

10

Expl 80%RH Expl 96%RH Pred 80%RH Pred 96%RH Pred 80%RH (Res) Pred 96%RH (Res)

0 0 10 20 30 Time (week) 40 50

Fig. 20. Prediction of the DLJ residual strengths (plane strain).

included but the reduction in strength was not signicant. When modelling the adhesive layer with an elastic continuum, which is not reported here, there was a reduction in strength of about 25% when the residual strains were included. The same characteristics can be seen for the wet joint (Fig. 22). The predictions were seen to be very good for the dry joint (Fig. 20). The results obtained from the 2D and 3D models gave nearly the same result for the dry condition when neglecting residual strains. When including residual strains, the 2D (plane strain) model gave a slightly lower failure load than the 3D model as the stress state in the 3D model was closer to plane stress for the thermal residual strains (Fig. 17). Plane strain was however used in the 2D model as this assumption was seen to accurately predict the stresses due to the mechanical load (Fig. 16).

Load (kN)

ARTICLE IN PRESS
C.D.M. Liljedahl et al. / International Journal of Adhesion & Adhesives 27 (2007) 505518 517

25

20

15

10
Expl 96%RH Pred 96%RH (2D) Pred 96%RH (3D) Pred 96%RH (Res, 2D) Pred 96%RH (Res, 3D)

0 0 10 20 30 Time (week) 40 50

Fig. 23. Comparisons between 2D and 3D (96%RH) when predicting the failure load.

However, for the wet joints the strength was overestimated. Thus modelling in 3D for joints aged in the 96%RH environment was undertaken to investigate if the diffusion from the sides contributed signicantly to the reduction in the strength. The results can be seen in Fig. 23. When the residual strains were neglected the prediction of the joints using the 2D model was slightly higher than for the 3D model, which was expected as moisture now also entered from the sides. When including residual strains the stresses did increase at the edge due to composite swelling but were lower in average throughout the width than in the plane strain model (Fig. 18) and did not reduce the strength as they were not large enough to shift the peak stresses from the compression side to the tension side. Hence, neither the diffusion from the side nor the swelling of the composite in the width direction was seen to reduce the strength signicantly more than when modelling in 2D. None of the models predicted the extent of degradation observed experimentally (Figs. 20 and 23). This may be due to composite failure, stress-enhanced diffusion or stressenhanced degradation. Composite failure could have been included in the model if the required experimental data were available as explained for the composite SLJ. Stressenhanced diffusion and stress-enhanced degradation have not been included for the same reasons. An approach to include these mechanisms has been discussed and applied to loaded aluminium SLJs elsewhere [16].

by immersion in de-ionised water were good. The degradation of the SLJ immersed in tap water may have been governed by cathodic delamination. The prediction of the composite SLJ was good for the unaged joints and for aging times up to 26 weeks in the 96%RH and 50 1C. The prediction for the joint aged at 80%RH and 70 1C were however overestimated. This was possibly due to composite failure, which was not included in the modelling. The prediction of the unaged DLJ was good. However, the 2D predicted strength of the aged joints were overestimated. The joint was therefore modelled in 3D to investigate if either the moisture penetration from the exposed edges furthest apart or the stresses due to swelling in the width direction reduced the predicted strength signicantly. This was however not the case. There are large residual stresses in the DLJ, which may have enhanced the degradation during ageing. The main aspects of future work are to develop a methodology to derive real CZM parameters for the composite failure from simple tests; to further investigate stress-enhanced degradation and to further investigate the effect of corrosion on the durability of adhesively bonded joints. This will require both extensive experimental testing and development of unifying models where all degrading mechanisms are included. Acknowledgements The authors gratefully acknowledge QinetiQ for manufacturing of the lap shear joints and composite laminates and for testing most of the aluminium SLJs and the composite laminates. Ms Y. Hua at the University of Surry is acknowledged for the testing of the DLJ and Mr. F. Jumbo at Loughborough for testing of the composite SLJs. J.P. Sargent at BAE is acknowledged for the moisture uptake and swelling coefcients for the composite and for testing of the thin SLJs. Thanks are also due to Prof. J. Watts for help with interpretation of the surface analytical studies and to Mr S. Greaves and Dr. S. Hinter for their assistance with the XPS analysis (all at University of Surrey). Further, the authors are grateful to MoD for nancial support. References
[1] Yu J-H, Gou S, Dillard DA. J Adhes Sci Technol 2003;17:149. [2] Loh WK, Crocombe AD, Wahab MMA, Ashcroft IA. Int J Adhes Adhes 2005;25:1. [3] Peretz D, Weitsman Y. J Rheol 1983;27:97. [4] Romanko J, Knauss WG. Technical report AFWAL-TR804037. [5] Xiao GZ, Shanahan MER. Polymer 1998;39:3253. [6] Cabanelas JC, Prolongo SG, Serrano B, Bravo J, Baselga J. J Mater Process Technol 2003;143:311. [7] Jurf RA, Vinson JR. J Mater Sci 1985;20:2979. [8] Reedy ED, Guess TR. J Adhes Sci Technol 1996;10:33. [9] Vaddadi P, Nakamura T, Singh RP. Compos A Appl S 2003;34:719. [10] Liljedahl CDM, Crocombe AD, Wahab MA, Ashcroft IA. J Adhes Sci Technol 2005;19:525.

4. Conclusions and future work The CZM parameters used in the prediction were determined from fracture test specimens, saturated in various relative humidity environments. The determined parameters have been used to predict the residual strengths of a number of different joint congurations. The predictions of the aluminium SLJ aged at 96%RH and

Load (kN)

ARTICLE IN PRESS
518 [11] [12] [13] [14] C.D.M. Liljedahl et al. / International Journal of Adhesion & Adhesives 27 (2007) 505518 [18] Krueger R, Paris IL, 0Brien TK, Minguet PJ. NASA/TM-2001210842, ARL-TR-2432. [19] Schon J, Nyman T, Blom A, A H. Compos Sci Technol 2000;60:173. [20] Sargent JP. Int J Adhes Adhes 2005;25:247. [21] Deorian F, Rossi S. J Adhes Sci Technol 2003;17:291. [22] Davis SJ, Watts JF. J Mater Chem 1996;6:479. [23] Watts JF, Castle JE. J Mater Sci 1984;19:2259. [24] Kinloch AJ, Korenberg CF, Tan KT, Watts JF. 7th EURADH, Freiburg, Germany, 2004. [25] Rattana A, Hermes JD, Abel ML, Watts JF. Int J Adhes Adhes 2002;22:205. Gledhill RA, Kinloch AJ. J Adhes 1974;6:315. Gledhill RA, Kinloch AJ, Shaw SJ. J Adhes 1980;11:315. Crocombe AD. Int J Adhes Adhes 1997;17:379. Loh WK, Crocombe AD, Wahab MMA, Ashcroft IA. J Adhes 2003;79:1135. [15] Crocombe AD, Hua YX, Loh WK, Wahab MA, Ashcroft IA. Int J Adhes Adhes 2006;26:325. [16] Liljedahl CDM, Crocombe AD, Wahab MA, Ashcroft IA. J Adhes, accepted. [17] Liljedahl CDM, Crocombe AD, Wahab MA, Ashcroft IA. Int J Fract 2006;141:147.

You might also like