Composites: Part A 33 (2002) 277288

www.elsevier.com/locate/compositesa

Residual stress development during the composite patch bonding process: measurement and modeling
D. Djokic a, A. Johnston b, A. Rogers a, P. Lee-Sullivan a,*, N. Mrad b
b

Department of Mechanical Engineering, University of New Brunswick, P.O. Box 4400, Fredericton, NB, Canada E3B 5A3 National Research Council Canada, Institute for Aerospace Research, 1500 Montreal Road, Building M-3, Ottawa, Ont., Canada K1A 0R6 Accepted 2 May 2001

Abstract This paper presents an experimental technique for monitoring residual stress development throughout the composite patch repair curing process. Using this technique, process-induced strains and specimen warpage during a number of different cure cycles were measured for a simulated single-sided composite patch repair of an aluminum substrate. Models for adhesive cure rate and glass transition behavior of the patch adhesive resin (FM 300-1K) were combined with a simple bi-metallic strip model to predict specimen warpage and strain behavior during cure. Model predictions were compared with experimental measurements and were used to assist in the development of optimized cure cycles. Using these optimized cycles, it was found that it was possible to achieve signicant (.20%) reductions in patch warpage and at the same time, minimize processing time and obtain a high nal adhesive degree of cure. Experimental observations suggest that an improved patch model incorporating adhesive viscoelastic behavior during cure would assist in achieving additional process improvements. q 2001 Elsevier Science Ltd. All rights reserved.
Keywords: A. Carbon bre; B. Residual/internal stress; C. Analytical modelling; D. Process monitoring; Patch repair processing

1. Introduction An increasingly important technology for life extension of aging aircraft is the use of bonded polymer composite patches for repair and/or local strengthening of primary and (more commonly) secondary aircraft structure. Compared with alternatives such as mechanically fastened metallic or composite patches, bonded composite patches provide both improved performance and weight savings, and for many applications represent the only viable life-extension approach. Although superior to most traditional repair methods, composite patch repair also has some drawbacks. One common problem arises due to the very different thermal expansion coefcients of typical composite patch materials and the metallic substrates to which they are often bonded. Since the cure process for the patch lm adhesive (and the patch itself, if not pre-cured) often takes place at high temperature, large thermal strain differences are observed between the patch and substrate materials. Reduction and control of the arising residual stresses is an important issue in ensuring maximum bonded repair performance. Depend* Corresponding author. Fax: 11-506-453-5025. E-mail address: [email protected] (P. Lee-Sullivan).

ing on the local stiffness of the underlying structure, these stresses may remain locked in or be relieved by patch and structure deformation. While structural deformation may result in changes in local aerodynamic characteristics, locked-in stresses create other difculties. Bondline residual stresses, for example, can result in premature failure of the patch/substrate bond. Residual stresses in the repaired structure tend to be tensile in nature. For cracked structures, the result is an increase in the crack tip stress intensity factor, KI; for the majority of other aircraft structures, fatigue life and load-carrying capacity are reduced. The result of process-induced residual stresses is thus a reduction in both patch efciency and repair durability [1,2]. A wide range of variations of the basic repair procedure are commonly used. In many cases, patches are applied either completely uncured or in a `B-staged' state, meaning partially cured, generally just past the point of thermoset matrix gelation. For such cases, nal cure of the composite patch is achieved in situ at the same time as the adhesive [3]. While this approach allows the composite patch to reshape itself somewhat to conform to the underlying substrate, both the patch and adhesive materials must remain refrigerated until use. Care must also be taken to match patch and adhesive cure cycles. Thus, if the area being patched is geometrically simple, the patch is often fully `pre-cured' prior

1359-835X/02/$ - see front matter q 2001 Elsevier Science Ltd. All rights reserved. PII: S 1359-835 X(01)00 083-5

278

D. Djokic et al. / Composites: Part A 33 (2002) 277288

to installation. This paper presents work performed in the rst stage of a research program focusing on this simplied patching process. Using this simplied approach, the investigation of internal strain development can focus exclusively on the curing adhesive. Once a better understanding of this process is obtained, the more complex case of B-staged patch repair will be studied. A common approach for minimizing process-induced patch stresses is to lower the patch curing temperature and increase the cure time, thereby reducing the temperature range over which thermal strains can develop [4,5]. However, if not used carefully, this approach can also result in a lower nal adhesive degree of cure and thus negatively impacts in-service patch performance. One objective for the current work was to conduct a preliminary examination of the use of reduced-temperature cure cycles for the reduction of residual stress during patch repair. The work also studied the evolution of adhesive properties during the cure process in order to develop optimized cure cycles to achieve a combination of minimum residual stresses and cycle time, and maximum level of cure in the patch adhesive. 2. Experimental approach 2.1. Experimental set-up Stress development during the composite patch repair process was monitored using the simulated single-sided patch repair specimen illustrated in Fig. 1. As shown, this specimen consists of an AS4/3501-6 composite patch

bonded to a at 2024-T3 aluminum substrate using a single ply of FM300-1K. An asymmetric specimen conguration was selected to produce large one-dimensional specimen warpage that could easily be measured. Large warpage was also ensured by the choice of thermally dissimilar materials and a high processing temperature adhesive. 2.2. Preparation of surfaces The 102 mm 50:8 mm patches were machined from a large (305 mm 305 mm) 10-ply ([0]10) AS4/3501-6 laminate plate. A single ply of FM300-1K adhesive lm was cocured to the bottom (bonding surface) of the plate to assist patch surface preparation and enhance subsequent patch substrate bonding. During later grit-blasting, most of the adhesive lm layer was removed, leaving only a very thin residual layer of cured adhesive that would become the bonding surface. The aluminum substrate required a more complex surface preparation procedure, conducted in three stages: mechanical preparation; chemical surface activation; and primer application. The procedure described by Albat [1] was used rather than a similar but more expensive treatment recommended by the adhesive manufacturer [6]. The rst step in the procedure consisted of removing the aluminum oxide surface by abrasion with SiC paper. The exposed surface was rinsed in water, air-dried and cleaned with methyl ethyl ketone (MEK). It was then grit blasted again to expose a clean bonding surface and a silane coupling agent was immediately applied to minimize surface oxidation. Finally, corrosion-inhibiting BR 127 primer was

Fig. 1. Schematic of patch specimen design (not to scale).

D. Djokic et al. / Composites: Part A 33 (2002) 277288

279

Fig. 2. Vacuum bagging of patch specimen.

sprayed onto the bonding surfaces followed by air-drying and oven curing. 2.3. Patch specimen assembly The specimen was assembled in a series of steps, as follows. First, a single layer of uncured FM300-1K adhesive was placed on top of the aluminum substrate surface, followed by the cured carbon-ber composite patch. A thermocouple was mounted onto the patch top surface, adjacent to a bonded foil strain gauge (Type CEA-13-250UW-350 [7]) labeled SG2 in Fig. 1. A second foil gauge was bonded to the opposite side of the specimen on the bottom surface of the aluminum substrate (SG3) using high temperature adhesive [8]. Finally, the composite patch assembly was covered with a layer of glass bleeder fabric, and two layers of polyester breather fabric. Strain gauges SG2 and SG3 were used to monitor specimen strain throughout the cure cycle. A `far-eld' strain gauge (SG1) was initially used to provide temperature compensation, but it was later determined that other techniques were required for adequate measurement accuracy

over the large temperature range encountered during processing (roughly 201808C) [9]. All strain gauges were connected to a computer-based data-acquisition system (DAQ) via a three-wire connection to provide thermal compensation. For curing, a bagging arrangement was used in which only the top of the patched section was actually under the vacuum bag. This was done in order to ensure that the entire specimen was as free as possible to deform during curing. The vacuum bagging procedure included sealing along the periphery of the patch area and extending the bagging for at least 20 cm from the patched area before attaching the vacuum connector. The patch assembly and vacuum bagging is shown in Fig. 2. The assembled patch specimen was mounted into a custom-made xture as illustrated in Fig. 3. This xture was designed such that the specimen was simply supported at the edges and completely free to expand, contract, and bend out-of-plane. Movement in the plane of the patch due to air movement in the oven (or in an autoclave, as will be used in subsequent experiments) was also effectively restrained. The xture was built to accommodate an instrumented cantilever beam mounted beneath the specimen,

Fig. 3. Fixture for mounting test specimen and cantilever beam.

280

D. Djokic et al. / Composites: Part A 33 (2002) 277288 Table 1 Modeling parameters Cure kinetics Parameter m E1 (J mol 21) A1 (s 21) E2 (J mol 21) A2 (s 21) a f1 (K) a f2 n1 (K) n2 Value 1.94 1.30 10 5 7.16 10 4 3.10 10 12 3.30 10 6 22.54 10 2 5.46 10 21 22.21 10 3 9.25 Glass transition temperature Parameter A1 B1 C1 A2 B2 C2 ac Value 1.40 10 1 5.46 10 1 8.38 10 1 1.21 10 2 3.50 10 2 24.13 10 2 0.844

used for in situ measurement of the specimen out-of-plane bending (warpage). To measure warpage, a pin attached to the free end of the cantilever was placed directly in contact with the underside of the specimen at its center point. The cantilever was then preloaded to ensure that contact was always maintained as the specimen deected during processing. The amount of deection was determined using strain gauges mounted at the root of the cantilever on its top and bottom surfaces. The strain gauges were connected in a halfbridge conguration for improved accuracy. To simulate the patch repair process, the entire test xture holding the vacuum-bagged specimen was placed inside a convection oven and subjected to various temperature cycles. During each cure process, specimen strains, cantilever deection and temperatures were measured using a computer-controlled (DAQ) system. After each cure cycle was complete, the maximum specimen warpage was veried using a micrometer. It should be noted that monitoring of patch and substrate strains was carried out concurrently with warpage measurements in a number of preliminary tests to verify the accuracy of cantilever measurements. Once veried, subsequent tests could be condently performed without the need for strain monitoring, saving a great deal of experimental effort. 2.4. Adhesive lm characterization A differential scanning calorimeter (TA Instruments DSC 2010) was used to characterize the cure rate behavior of the FM300-1K adhesive resin and the development of glass transition temperature (Tg) with the progression of cure. DSC scans at various isothermal cure temperatures and dynamic scan rates were rst conducted to develop a kinetics model to predict the degree of cure as a function of time and temperature. A modied version of an empirical cure kinetics model of the type employed by Kamal and Sourour [10], shown in Eqs. (1) and (2) was used: da K1 1 K2 am af 2 an dt  2DE1 K1 A1 exp ; RT  2DE2 K2 A2 exp RT   2 1

After determining the a f vs. T relationship, model rate constants K1 and K2, pre-exponential factors A1 and A2, activation energies DE1 and DE2 and exponents m and n were calculated using a least-squares optimization routine. While the parameter m was found to remain essentially constant with temperature, n was highly temperature-dependent with a dependency described by a relation of the form of Eq. (4):   n1 n exp 1 n2 4 T where n1 and n2 are model constants and T is the resin cure temperature. Calculated cure kinetics model constants are provided in Table 1. As shown in Fig. 4, very good agreement was found between model predictions and isothermal DSC measurements. The relationship between glass transition temperature and degree of cure was then determined. This was done by rst isothermally curing a number of adhesive specimens to various extents, then cooling them to room temperature. For each specimen, Tg and a were determined during a single dynamic DSC scan, the former from the inection point of the endothermic portion of the heat ow curve, and the latter by measuring the residual heat of cure and comparing to the total heat of reaction of uncured material.

where a is the degree of cure, and t, T and R are time, absolute temperature and the universal gas constant, respectively, and DE1 and DE2 are activation energies. The maximum attainable degree of cure at a given isothermal temperature, a f, was found to vary with temperature as described by Eq. (3):   af1 1 af2 af exp T , 465:2 K 3 T

af 1:0

T $ 465:2 K
Fig. 4. Measured and predicted degree of cure (a ) as a function of time for FM 300-1K adhesive at various isotherms.

where af 1 and af 2 are model constants and T is the resin cure temperature.

D. Djokic et al. / Composites: Part A 33 (2002) 277288

281

Fig. 5. Measured and predicted change in glass transition temperature as a function of measured a at 1508C cure.

It was found that a two-part second order polynomial model, as shown in Eq. (5), was sufcient to describe the relationship between glass transition temperature and degree of cure: Tg A1 1 B1 a 1 C1 a2

d1 r1 2 cos Q d 2 A sin Q

a , ac a . ac

Tg A2 1 B2 a 2 ac 1 C2 a 2 ac 2

The parameters A1, B1, C1, A2, B2 and C2, and a c were calculated using a least-squares t to the experimental measurements and are shown in Table 1. Fig. 5 shows the comparison between predicted and measured Tg when curing the adhesive at 1508C. 2.5. Predictions of specimen deformation Predictions of patch and substrate strains and specimen deection during processing were derived from a closedform solution for a bi-metallic strip by Timoshenko [11]. The total deection at the center of the specimen, d , was calculated from the sum of the patched section deection, d 1, and the deection of the specimen `free arm' section, d 2, (see Fig. 6) as in Eqs. (6) and (7):

d d1 1 d 2

where r and Q denote, respectively, the radius of curvature and included angle of the section under the patch, and A is the length of the `free arm section'. Using this model to calculate specimen deections requires knowledge of patch and substrate thermal strains. In many cases, it would be sufcient to assume a constant coefcient of thermal expansion (CTE) for both the patch and substrate, obtained from the measurements of the bonded strain gauges. However, preliminary tests on stand-alone (unbonded) patches and aluminum substrates showed that the measured CTEs over the range of interest to be quite non-linear. Signicant variations were observed in thermal strain measurements of even nominally identical strain gauges (from a single batch) bonded to identical locations. To minimize this variation as a source of error, strain gauge measurements taken during the heat-up phase of each test were used to calculate coefcients of thermal expansions for the patch (CTEP) and substrate (CTES). These coefcients were then t using third

Fig. 6. Schematic of warped patch specimen due to out-of-plane bending.

282

D. Djokic et al. / Composites: Part A 33 (2002) 277288

order polynomial equations as: CTE P A3 1 B3 T 1 C3 T 1 D3 T

CTES A4 1 B4 T 1 C4 T 2 1 D4 T 3 For a given temperature change, patch and substrate thermal strains were then calculated using: Z T2 1P CTEP dT 9
T1

1S

Z T2
T1

CTES dT

single-step cycles, the vast majority of cure occurs at the maximum cure temperature, in multi-stage cycles, curing occurs over a wide range of temperatures. Previous work such as that of Cho and Sun [2] has shown that in such cases no simple relationship between cure temperature and stress free temperature exists. In this work, the specimen warpage model (Eqs. (6)(9)) was used to calculate an effective stress free temperature, TSF. This was done for each specimen by starting with the measured room-temperature deformation and modeling its change in shape with increasing temperature. The temperature at which the specimen was predicted to become at was taken as TSF. 2.7. Process cure cycles As outlined in Table 2, residual stress development during ve different sets of process cycles was measured as part of the current work. In the rst set (Tests 13), the effect of maximum curing temperature was examined by curing the patch adhesive using three simple manufacturer-recommended cure cycles [6] with different `hold' temperatures: 1778C (Test 1), 1508C (Test 2) and 1208C (Test 3). The second set of cure cycles (Tests 46) were the same as the rst except that a lower cooling rate was used at the end of the cure process (i.e. 0.58C min 21 as opposed to 2.58C min 21 used in all other tests). White and Hahn [12] reported some success in using this approach without adversely affecting nal properties, although for a different type of thermosetting system. The objective of the third set of tests (Tests 79) was to examine the effect of post-curing the specimens to achieve a higher nal degree of cure (a ), as has been suggested in Ref. [1]. For these tests, maximum warpage was measured as the specimens from Tests 13 were re-heated briey (for 5 min)

where 1 P and 1 S are modeled patch and substrate strains, respectively. Temperatures T1 and T2 dene the limits of integration. In all cases, a simple viscouselastic constitutive model was employed in which the adhesive was assumed to exhibit viscous behavior prior to gelation and elastic behavior thereafter. Thus, no interaction was assumed between the patch and substrate until gelation and perfect bi-metallic strip behavior was assumed thereafter. In all cases, the pre-cured patch and substrate were assumed to exhibit fully elastic behavior throughout the process. 2.6. Stress free temperature A geometry-independent measure often used to quantify levels of process-induced residual stress is the `stress free temperature', TSF. This is the temperature at which a component (in this case the patched specimen) would reach a stress-free state, upon re-heating at the end of the cure process. In simple single-step cure cycles, a common and often reasonably accurate assumption is that TSF is equal to the maximum curing temperature. For complex, multi-stage cure cycles, however, the situation is not so simple. While in
Table 2 Experimental test matrix Tests Set 1 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Specimen S1-300 S12-300 S14-300 S2-300 S3-300 S4-300 S1-300 S3-300 S4-300 S6-300 S9-300 S8-300 S15-300 S16-300 S17-300 Process cycle type Single-Step Single-step Single-step Single-step Single-step Single-step Single-step (post-cure) Single-step (post-cure) Single-step (post-cure) Two-step Two-step Multi-step (short) Multi -step (short) Multi -step (long) Multi-step (long)

Hold temperature(s), Tc (8C) 177 150 120 177 150 120 177 177 177 150/177 120/177 150/130/177 150/130/177 150/130/177 150/130/177

Hold time(s), th (min) 60 160 480 60 160 480 5 5 5 60/30 330/30 28/40/38 28/40/38 28/55/38 28/55/38

Cool-down rate, r (8C min 21) 2.5 2.5 2.5 0.5 0.5 0.5 2.5 2.5 2.5 2.5 2.5 2.5 2.5 2.5 2.5

Set 2

Set 3

Set 4 Set 5

D. Djokic et al. / Composites: Part A 33 (2002) 277288 Table 3 Predictions from analytical models Tests Total processing time, t (h) 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 3.1 4.7 9.3 7.3 7.8 11.9 2.2 2.2 2.2 3.9 8.7 5.8 5.8 6 6 Degree of cure, a 0.98 0.94 0.86 0.98 0.94 0.86 0.98 0.96 0.96 0.98 0.98 0.98 0.98 0.98 0.98 Glass transition temperature, Tg (8C) 161 149 128 161 149 128 161 156 155 160 160 161 161 161 161

283

Set 1

Set 2

Set 3

Set 4 Set 5

than the single-step cycles (Tests 2 and 3), while also reducing the total processing time and maintaining low stressfree temperature. The fth and nal set of tests (Tests 1215) was developed as an extension of the two-step optimized cure cycles. Two variations of more complex, multi-stage tests were evaluated. The goal was to further minimize residual strains by designing the path of the Tg and a development. These cure cycles were designed to achieve an optimum compromise between cycle time, nal adhesive degree of cure and nal warpage. Predicted values of adhesive Tg and a at the completion of each process cycle were used to evaluate whether an adequate level of cure had been achieved. Although it is understood that `adequate' varies depending on the application, values of Tg 1508C and a 0:95 were chosen here as a benchmark. Predicted values for each of these parameters at the end of all test cycles are presented in Table 3. 3. Results and discussion 3.1. Effect of cure temperature (Tests 13) As indicated in Table 5, Tests 13 verify that patch residual stress (as indicated by specimen warpage) can indeed be reduced by lowering the cure temperature. For example, maximum specimen warpage at room temperature was found to decrease from 5.6 to 4.0 mm when cure temperature was lowered from 177 (Test 1) to 1208C (Test 3). A corresponding reduction in stress free temperature, TSF, from 178 to 1228C was observed. However, this reduction in residual strain was achieved at the expense of an increase

to the maximum curing temperature (1778C) and then cooled back to room temperature at the normal rate of 2.58C min 21. The fourth test set (Tests 10 and 11) examined the possibility of using the developed models for a and Tg to devise more optimized `two-step' cure cycles. Tests 10 and 11 employed cure cycles consisting of a low rst hold temperature, at 120 and 1508C, respectively, followed by a nal cure at 1778C. The intention was to achieve substantially higher values of degree of cure and glass transition temperature
Table 4 Specimen warpage results Tests Final warpage, d (mm) Micrometer measurements Set 1 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 5.6 4.5 4.0 5.4 5.0 4.0 5.6 5.4 5.2 5.0 5.0 5.1 4.7 4.5 4.4

Cantilever (strain gauge) measurements 5.4 4.8 3.6 5.3 5.0 3.9 5.4 5.8 5.7 5.1 5.0 5.3 4.7 5.0 4.6

Stress free temperature, TSF (8C)

178 150 122 178 152 123 178 172 169 165 167 168 158 153 151

Set 2

Set 3

Set 4 Set 5

284 Table 5 Residual strain results Tests Exp. Set 1 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Patch strain, 1 P (m1) Modeled 311 N/A N/A 386 271 277 N/A N/A N/A 245 154 314 N/A N/A N/A Exp. 175 141 109 165 99 170 175 213 158 250 124 173 68 80 185

D. Djokic et al. / Composites: Part A 33 (2002) 277288

ineffective in lowering of the residual strain and did not justify the longer processing times required (2257%).
Substrate strain, 1 S (m1) Modeled 21004 N/A N/A 2984 2950 2684 N/A N/A N/A 21007 2963 2944 N/A N/A N/A 21000 2809 2622 21005 2862 2651 21000 2838 2638 2866 2676 2691 2643 2665 2665

3.3. Effect of post-cure (Tests 79) As mentioned, other researchers have suggested postcuring as a remedy for the low nal Tg and a produced by low-temperature cure processes. As shown in Table 3, the brief (5 min) post-cure conducted in Tests 79 did indeed increase nal degree of cure. However, these improvements came at the cost of negating much of the warpage reduction obtained using the original reducedtemperature cure. The source of the measured increase in warpage can be deduced from Fig. 8, which shows the change in deection of specimens initially cured at 120, 150 and 1778C, as they are heated for post-cure. In this gure, the initial warpage indicated for each specimen is re-set to zero so that the curves for each will overlap near room temperature. During the initial stages of the post-cure cycle, all three specimens exhibit identical, approximately linear, changes in deection as they begin to atten. As temperature continues to increase, the deection for all specimens diverge from linearity well before reaching the post-cure temperature of 1778C. The developed glass transition model indicates that the point of divergence for each specimen occurs very near the adhesive Tg attained after the initial cure cycle (see Tests 16). When specimen temperature exceeds this initial Tg, adhesive stiffness decreases signicantly and additional stresses that develop with increasing temperature mostly relax. During post-cure, the cure level in each specimen increases and they all ultimately achieve nearly the same Tg (that is, 1551618C). As a result, during cool-down, the three specimens follow nearly parallel deection vs. temperature paths. The result of the observed asymmetry in heating and cooling behavior is an increase in the warpage (and TSF) of the specimen initially cured at 1208C of about 1.2 mm

Set 2

Set 3

Set 4 Set 5

in process cycle time from 3.1 to 9.3 h. Perhaps more important from the standpoint of patch performance, the predicted maximum adhesive resin degree of cure decreased from 0.98 to 0.86 and the predicted maximum Tg was reduced from 161 to 1288C (Table 3). 3.2. Effect of cooling rate (Tests 46) As shown in Table 3, using a slower cool-down rate of 0.58C min 21 resulted in minimal reduction of warpage and TSF. Fig. 7 shows that specimens cured at 1778C and cooled at different rates exhibited almost identical warpage development behavior. In both instances, viscoelastic response was noted, as indicated by the non-linear deection vs. temperature curve. However, in the tested range (2.50.58C min 21), adhesive stress relaxation was evidently, not signicantly, inuenced by the cooling rate. Thus, the method in this case was

Fig. 7. Specimen deection due to cooling at 2.58C min 21 (standard) and at 0.58C min 21 (slow).

D. Djokic et al. / Composites: Part A 33 (2002) 277288

285

Fig. 8. Measured specimen deection during post-cure (Tests 79).

(548C) and of about 0.9 mm (268C) in the specimen initially cured at 1508C. No change in shape or TSF is observed in the specimen initially cured at 1778C. Thus, although post-cure improved nal degree of cure, as shown in Fig. 9, the improvement in this case came at the cost of an increase in warpage and TSF. The analytical model used in this work failed to predict this effect since the viscouselastic constitutive model used could not account for the viscoelastic response of the curing adhesive. However, it is apparent that the strategy of using a lowtemperature cure followed by high-temperature post-cure did result in some level of warpage reduction. This is believed to be due to the fact that, although the specimens mostly re-equilibrated at the post-cure temperature, some of the stresses developed when increasing from the initial cure temperature to the post-cure temperature did not relax. This phenomenon was exploited when developing the `optimized' cure cycles employed in Tests 10 and 11.

3.4. Effect of two-step hold cycle (Tests 10 and 11) In Tests 10 and 11, resin cure rate and Tg development models were exploited to predict how the cure cycle could be modied in order to minimize stress relaxation while increasing from a lower temperature `hold' to a higher post-cure temperature. The strategy used in these tests was to minimize the difference between the adhesive temperature and its predicted Tg throughout the temperature `ramp' from the low hold temperature to the post-cure temperature (i.e. minimize DTg T 2 Tg ). Although an innitely slow ramp rate would obviously have resulted in the lowest possible DTg, a target value of 168C was chosen. This value represented the approximate difference between the postcure temperature of 1778C and the `ultimate Tg' (i.e. Tg at a 1:0) of about 1618C, and was chosen to balance minimum cycle time and minimum residual stress objectives. The process prole designed by this method (Test 10) is illustrated in Fig. 10, which also shows the development of

Fig. 9. Effect of post-cure on a and stress free temperature, TSF.

286

D. Djokic et al. / Composites: Part A 33 (2002) 277288

Fig. 10. Prole of the two-step cure cycle used in Test 10.

a and Tg. This cure cycle employed an initial hold temperature of 1508C, maintained for 60 min, followed by a hold at 1778C for 30 min. In order to minimize DTg, the specimen was heated at 1.08C min 21 between the two hold temperatures, as compared to the normal rate of 2.58C min 21 used in standard cure cycles. As shown in Table 4, this strategy proved quite successful. While the nal maximum warpage remained in the same range as measured in Tests 2 and 5 (which also used a 1508C hold temperature), the cycle time was 0.5 h shorter and, more importantly, the nal degree of cure was increased from 0.94 to 0.98. Final warpage for this test was 0.4 mm less than Test 8 (initial 1508C cure 1 post-cure), and total cycle time was 2.7 h shorter. This indicates that, while crude, there may be some merit to the optimization approach employed here. However, the need for some caution in this assessment is indicated by the results of Test 11, which used a 1208C initial hold temperature. In this case, while the cycle time

and warpage were improved as compared to Test 9 (initial 1208C cure 1 post-cure), the warpage was much greater than that for Tests 3 and 6 (1208C cure only). The poor results are thought to be at least partly attributable to the fact that the optimization was based on a simplied kinetics model which did not account for diffusion, which becomes important at the later stages of the initial temperature hold [13]. 3.5. Effect of three-stage hold cycle (Tests 1215) The nal set of tests (Tests 1215 in Table 2) consisted of a series of four three-stage processing cycles similar to that illustrated in Fig. 11. During these tests, the specimen temperature was rst increased at the normal rate of 2.58C min 21 to 1508C. At this temperature, adhesive degree of cure and Tg increased at a fairly rapid, yet not excessive rate. After 28 min at 1508C, specimen temperature was decreased at 28C min 21 to 1308C where it was maintained

Fig. 11. Prole of the three-step cure cycle used in Test 12. The second hold is 55 min long.

D. Djokic et al. / Composites: Part A 33 (2002) 277288

287

Fig. 12. Comparison of nal degree of cure (a ) and stress free temperature (TSF) for single-step, two-step and three-step cure cycles.

for sufcient time for vitrication to be approached (i.e. DTg approached a `small enough' value). Next, the adhesive was heated to 1778C at a rate sufciently low (0.58C min 21) so as to ensure that a small and nearly constant DTg was again maintained until late in the cure process. The concept behind minimization of DTg was again that relaxation times would be long enough to prevent relaxation of much of the stress developed during heating from 130 to 1778C. Thus, the structure shape would be effectively `locked in' at the lowest possible temperature, in this case somewhere between 130 and 1778C. Preliminary optimization trials revealed that adhesive Tg is quite sensitive to the process time and temperature variations during the 1308C hold. To address these concerns, two tests for each of two variations of this three-step cycle were conducted, as shown in Table 2. The results presented in Table 4 indicate that this optimization method was effective in achieving further reductions in residual stress levels. The slightly larger than expected specimen deformation in Test 12 is attributed to an insufcient level of cure during the 1308C hold. Reviewing the actual process prole showed that poor process control had resulted in a lower than expected cure temperature in this process segment, as well as a premature ramp to 1778C. More stringent process cycle control in Test 13 yielded improvements in warpage and stress free temperature. A slightly more robust version (i.e. less prone to process variations) of this cure cycle was investigated in Tests 14 and 15. In these tests, the 1308C hold was extended from 40 to 55 min, to very near the predicted point of vitrication. As shown in Table 4, these process cycles resulted in even more reduction in part warpage. For these specimens, measured stress free temperatures (151 and 1538C) were essentially equivalent to those from the moderate temperature single-stage cycles (151 and 1528C). Of course, the optimized cycles also achieved improved nal Tg and degree of cure.

All complex cycles required an approximately 20% increase in processing time, due to the necessity for some curing at low temperature. In many cases, this would be considered a worthwhile compromise given the concomitant improvements in nal a and Tg, and reductions in thermal strains. The progress of this work from standard, through two-step and to complex cure cycles is shown in Fig. 12. As mentioned, only the warpage results were used in residual strain assessment. Patch and substrate strain data validated the in situ warpage measurements. The actual and predicted strains are presented in Table 5. 4. Conclusions This paper presents a simple approach for minimizing residual stress development during the processing of adhesively bonded composite patch repairs. The experiments conducted show that simply lowering the adhesive resin curing temperature reduced the process-induced residual stress (as manifested in the degree of specimen warpage), but that this came at the cost of much longer cycle times and reduced resin degree of cure and glass transition temperature. It was also found that, while post-curing addressed the problem of under-cure, viscoelastic stress relaxation during this process acted to negate much of the benet of the initial low-temperature cure. The predictions of a bi-metallic strip analysis of the single-sided patch showed a good match with experimental measurements for one-step process cycles, but fared poorly in cases where viscoelastic behavior was found to be important. Using this analysis in conjunction with models for resin cure and glass transition temperature development, optimized processes were developed, which achieved some of the stress-reduction benets of low-temperature processes while both improving nal degree of cure and reducing cycle time. Improved approaches for optimization of patch

288

D. Djokic et al. / Composites: Part A 33 (2002) 277288 repairs on metallic aircraft structures. Int J Adhes Adhes 1992;12:1926. Deskin R. FM 300-1K adhesive lm material data sheet. West Paterson, NJ: Cytec Fiberite Incorporation, 1999. Micro-Measurements. Precision strain gauges CEA-13-250UW-350; Engineering data sheet, 1999. Micro-Measurements. Instruction bulletin B-129, Surface preparation for strain gauge bonding, 1999. Djokic D. An investigation of the development of residual thermal strains in composite patch repair of metallic structures, Senior year report. New Brunswick, Canada: Department of Mechanical Engineering, University of New Brunswick, 2000. Kamal MR, Sourour S. Kinetics and thermal characterization of thermoset cure. Polym Engng Sci 1973;13(1):5964. Timoshenko S. Strength of materials. 3rd ed. New York: Robert E. Krieger, 1976. White SR, Hahn T. Cure cycle optimization for the reduction of process-induced residual stresses in composite materials. J Compos Mater 1993;27(14):135278. Atarsia A, Boukhili R. Relationship between isothermal and dynamic cure of thermosets via the isoconversion representation. Polym Engng Sci 2000;40(3):60720.

repair curing processes are expected as enhanced analyses that directly incorporate models for adhesive viscoelastic response during cure are developed. References
[1] Albat AM. Thermal residual stresses in bonded composite repairs on cracked metal structures. PhD Thesis. Department of Mechanical Engineering, The University of British Columbia, Vancouver, BC, Canada, 1998. [2] Cho J, Sun CT. Lowering thermal residual stresses in bonded composite patch repairs. Proceedings of the 14th American Society for Composites Technical Conference. Dayton (OH, USA), 1999. p. 894903. [3] Raizenne D, Roohi N, Sova M, Brothers MR. Experimental verication of graphite composite three-sided patches for CF-18 IFOSTP FT55 Y470.5 Bulkhead X-19 location progress report 1, A report to national research council institute for aerospace research structures. Materials and Propulsion Laboratory, 1998. [4] Baker AA, Jones R, editors. Bonded repair of aircraft structures Netherlands: Martinus Nijhoff, 1988. [5] Ong CL, Shen S. The reinforcing effect of composite patch

[6] [7] [8] [9]

[10] [11] [12]

[13]