You are on page 1of 15

International Journal of Adhesion & Adhesives 22 (2002) 721

Through-thickness shape optimisation of bonded repairs and lap-joints


R.H. Kaye, M. Heller*
Airframes and Engines Division, DSTO, Aeronautical and Martime Research Laboratory, P.O. Box 4331, Melbourne, VIC 3001, Australia Accepted 11 October 2000

Abstract For realistic applications, the design of bonded repairs and lap-joints, has often been undertaken through trial and error nite element analyses, or experiment. Recent experience indicates that for more complex practical applications, unacceptably high adhesive stresses can occur in the adhesive layer. In the present work, an automated sensitivity-based shape optimisation procedure has been developed for the optimal design of free-form bonded repairs and lap-joints, with the aim of achieving reduced adhesive stresses. The approach has been demonstrated through application to a number of single and double-sided congurations where both the shapes of the adhesive layer and the outer adherend are allowed to vary. Signicant improvements over conventional designs are obtained, as assessed by the reduction in peak adhesive stresses. These results indicate that the numerical shape optimisation procedures presented can provide designs that oer substantial improvements over standard designs. Crown Copyright r 2002 Published by Elsevier Science Ltd. All rights reserved.
Keywords: C. Finite element stress analysis; C. Stress distribution; D. Fatigue; E. Joint design

1. Introduction 1.1. Bonded repairs and scope for optimisation Over approximately the last two decades, the use of high modulus bonded doublers has been demonstrated as an eective procedure for the life extension of cracked and uncracked metallic airframe components [1]. These repairs can be used either alone or in combination with shape reworking. The bonded repairs function by transferring some load from the component, through the adhesive layer into the repair, thereby reducing the magnitude of cyclic stresses in the repaired component. The thickness of the doubler (repair) determines how much load is drawn away from the damaged region and also the level of stress in the adhesive. Typically the doubler is a laminate consisting of unidirectional layers of boron bre embedded in a matrix of epoxy resin. This allows for thinner doublers due to the high bre
*Corresponding author. Tel.: +61-3-9626-7637; fax: +61-3-96267089. E-mail address: manfred.heller@dsto.defence.gov.au (M. Heller).

modulus and it is better suited to inspection by NDI methods than laminates consisting of other bres (i.e. carbon). Heat cured epoxy adhesives and rened methods of surface preparation have generally provided high strength and durable bonding. When applied to primary structural components, these repairs are typically used as a measure to prevent crack initiation or retard crack growth. It is generally required that the component has adequate static strength with or without the repair. To date, most applications of bonded repairs have been to thin sections such as cracked airframe skins (i.e. a plate). Here, the stress analysis has been usually based on an analogy with a one-dimensional lap-joint analysis, where 100% of the load is carried by the doublers [2,3]. A key quantity of interest is the adhesive stress concentrations at the extremities of load transfer regions [4,5]. These occur at (i) the end of the bonded repair, and (ii) directly over the crack. Often yielding of the adhesive can occur at these locations, and for extreme cases, this can lead to premature adhesive failure depending on the specic service loading history. For more complex repair applications, involving curved thick sections, and/or

0143-7496/02/$ - see front matter Crown Copyright r 2002 Published by Elsevier Science Ltd. All rights reserved. PII: S 0 1 4 3 - 7 4 9 6 ( 0 1 ) 0 0 0 2 9 - X

R.H. Kaye, M. Heller / International Journal of Adhesion & Adhesives 22 (2002) 721

involving signicant bending, a particular feature is the existence of both high shear and peel stresses. This can potentially seriously compromise the integrity of the adhesive layer, as there will always be a trade o between achieving a suitable repair eectiveness (i.e. load transfer), and restricting adhesive stresses to be less than an acceptable limit [6,7]. It should also be noted that practical applications to date have essentially used a constant adhesive thickness as well as a constant repair thickness (except for tapering at the ends of the repair). Typical designs have not been optimal, although various strategies have been used to reduce the magnitude of the stress concentration. These include linear tapering at the end of the joint, and/or increasing the local adhesive thickness at the end of the joint [8]. Generally, no closed form theoretical solutions are available for the design of more complex repairs, hence trial and error FE analyses or experiments have been used to obtain a suitable design. Published work on the optimal design of bonded repairs/lap-joints to reduce adhesive stress is very limited. However some investigations of specic scope have been undertaken. For example an analytical treatment of the optimal tapering at the ends of an isotropic doubler for a uniaxialy loaded lap-joint is given by Ojalvo [9]. In other work the same problem is considered by using a 2D nite element gradientless optimisation method, (one thick section repair is also considered) in Refs. [10,11]. In the work of Groth and Nordland [12] design sensitivity methods are used to optimise essentially the same conguration. In all four references above, the analyses are conned to the consideration of doubler tapering and do not consider variation in adhesive thickness. It should be noted that concurrent with the present investigation the authors have also recently completed a study for the throughthickness optimal design of a bonded repair, as a life extension option for the FS 470 bulkhead of the F/A-18 airframe [13]. Here sensitivity based nite element optimisation methods have been used and both the repair and the adhesive layer are allowed to vary in shape. 1.2. Present work In the present work, an automated sensitivity-based shape optimisation procedure has been developed for the optimal design of free-form bonded repairs, with the aim of achieving reduced adhesive stresses. The approach has been demonstrated through application to a number of single and double-sided congurations. The analyses undertaken are completely relevant to the design of lap-joints as well. Particular features of the present approach include: (i) free form shapes, where the outer adherend and/or the adhesive thicknesses are allowed to be non-uniform, and are

optimised, (ii) a least squares objective function is used to undertake the optimisation, subject to specied constraints, and (iii) multiple shape-basis vectors from the analysis of an auxiliary model are used to specify allowable shape changes. Four general loading and geometric cases were considered as follows; where in each case the inner adherend is aluminium and xed in geometry: (i) In case 1 the geometry representative of a symmetric (i.e. double sided) crack repair is modelled. Here the crack opening is prevented from increasing during the optimisation process; hence the repair eectiveness (as compared to a nominal repair) is maintained as a constant while the adhesive stresses are minimised. In this model, aluminium doublers are bonded to an equivalent stiness aluminium plate. (ii) Case 2 is the same as case 1 except that the doublers have been modelled using homogeneous orthotropic boron/epoxy properties. (iii) Case 3 is similar to case 2 except, the boron/ epoxy doubler is bonded to one side of the plate only. Hence due to the resulting oset of the neutral axis, the joint experiences signicant local bending. This is a challenging case since very high adhesive stresses can be envisaged. (iv) In case 4, the geometry is representative of a double-sided lap-joint rather than a crack repair where there is no restriction on joint opening. In this model aluminium doublers are bonded to an equivalent stiness aluminium plate. Here while the adhesive thickness is allowed to vary in the optimisation process, the outer surfaces of the doublers are constrained to remain at. Detailed denition of the congurations follows in later sections with corresponding gures. In all cases, the objective has been to minimise the peak von Mises stress by making the adhesive stress distribution more uniform. This is achieved by dening an objective function as the sum of the squares of the stress deviations from the average value, over a selected region. While the work has been presented as a numerical study without supporting experimental data it has been shown experimentally in [8] that reducing adhesive stress concentrations leads to greater joint strength.

2. Problem denition and analysis approach 2.1. Lap-joint loading and initial conguration For all analyses the initial conguration under study was a typical bonded double lap-joint as dened in Fig. 1. This conguration is perhaps more correctly termed a double strap joint but has been referred to as a double lap-joint in much of the literature. The geometry

R.H. Kaye, M. Heller / International Journal of Adhesion & Adhesives 22 (2002) 721

Fig. 1. Dimensions and loading arrangement for double lap-joint.

is also representative of the two dimensional idealisation of a bonded repair to a cracked plate [2,3]. In this case the two outer adherends represent the doublers, while the two inner adherends represent the two sections of a cracked plate. Consequently the terms joint and crack are both used to describe the centreline discontinuity of the base structure. For all cases, unless noted otherwise, the inner adherends have a thickness of 4 mm and are subjected to a remote uniaxial stress of 100 MPa. The outer adherends are bonded to the inner adherends by an adhesive layer having a nominal thickness of Z 0:15 mm. The outer adherends are both 120 mm in length and have a thickness of 2 mm if aluminium, or 0.67 mm if boron/epoxy. As is commonly advocated, an initial linear taper with a 1 : 10 slope was used at the ends of the outer adherends to reduce the magnitude of the adhesive stress concentration at the end of the doublers. For joint representations the geometry is considered to extend indenitely in the out-of-plane direction without change in shape or stress eld. For crack repair representations the geometry is considered to be a worst case 2D slice of unit thickness through the repaired plate, where the slice is (i) aligned with the remote applied stress direction, (ii) perpendicular to the crack, and (iii) is located in the centre of the plate [2]. Where boron/epoxy doublers are modelled, the bre direction is aligned to the direction of applied load. As indicated in Fig. 1, the joint has been divided into six load transfer regions i.e.: three on each side of the symmetry line x 60 mm, having an arbitrary length of 20 mm each. In the taper and joint/crack regions most of the load transfer takes place between the doublers and the plate, and this is where the shape changes were undertaken. It is common practice to have a region in between where there is no load transfer termed the separation region in this paper. A key reason for having this region is to provide a safety buer zone (i.e. potential load carrying region), should there be localised adhesive dis-bonding or voiding in the nominal load transfer regions. The presence of this region also helps

restrict relative movement between the inner adherend (plate) and doublers (in the loading direction), which may occur due to adhesive plasticity or creep in the nominal load carrying regions. Since in the separation region the adhesive has zero shear stress, the direct stress in the load direction is the only component present (for symmetric 2D analyses). The load and stress magnitudes in the doubler and plate are therefore dened by their relative stinesses. 2.2. General nite element modelling considerations In the present investigation linear elastic stress analyses were performed using the MSC/NASTRAN Version 70 linear static analysis nite element code processor running on a Hewlett-Packard K series 9000 computer at AMRL. The MSC/PATRAN level 7.5 code was used for pre and post processing of the models. Either quarter or half symmetry was used in the nite element analyses as appropriate, with plane strain conditions being used for all cases. The nite element meshes consisted of 4 noded rectangular elements in preference to 8 noded isoparimetric quadrilateral elements. This was done since the 4 noded elements (i.e. with two nodes per side) are more convenient to use with the NASTRAN 2 noded beam elements, which were an important part of the modelling for optimisation as explained in later sections. For all analyses the following material properties were used as appropriate: (i) for aluminium, isotropic material behaviour was assumed with Youngs Modulus, E 70; 000 MPa, and Poissons Ratio, v 0:35; (ii) the isotropic epoxy adhesive had Youngs Modulus E 840 MPa, and Poissons Ratio, v 0:3; and (iii) the unidirectional boron/epoxy adherand was taken as 2D orthotropic with E11 210; 000 MPa, E22 25400 MPa, v12 0:18; and G12 7200 MPa where the subscript 1 refers to the direction of the bres and 2 is the through thickness direction. Here the 1 and 2 directions are aligned with the x- and y-axes, respectively.

10

R.H. Kaye, M. Heller / International Journal of Adhesion & Adhesives 22 (2002) 721

2.3. Optimisation strategy Shape optimisation was achieved using the general sensitivity-based optimisation technique available in the MSC/NASTRAN code [14]. The implementation has been successfully developed in the present work and other concurrent investigations at AMRL, for both twodimensional rework optimisation to minimise peak stresses [7,13], and bonded repair design [13]. In this section the general approach used is summarised, while specic details for the particular conguration cases are presented in subsequent sections. In broad terms, the approach involves changing the position of nodes dening a boundary shape, based on the computation of rates of change of nodal stresses with respect to shape changes. Hence the general problem denition for shape optimisation can be stated as; nd values of: to minimise/maximise: subject to: design variables objective function constraint functions

The software allows for very wide scope as to how to dene the quantities above and preparing the problem cannot be considered a routine matter as may be the case with some other types of computational solutions. 2.3.1. Objective function and general considerations The objective function used in this work has been termed the least squares objective function in prior AMRL work [7,13]. It seeks to minimise the deviations from the average von Mises adhesive stress value in both the joint region and the taper region, and is given in Eq. (1). minimise fobj
i1 X in

savt si 2

j1 X jk

savj sj 2 ;

where i and j are the number of an individual adhesive element in the taper and joint regions, respectively, n and k are the total number of adhesive elements in the taper and joint regions, respectively, and savt and savj are the average adhesive element stresses in the taper and joint region, respectively. The von Mises stress term has been used so that peel stress is included in the process as well as shear. It would have been possible to do this using maximum principal stress also. The adhesive was modelled using three rows of elements and the stress values are taken at the centroids of centre row adhesive elements. All centre row elements in the taper and joint/crack regions were included except where noted otherwise in Sections 5.3 and 6.2. In the literature, for general structural optimisation, it appears common practice to use the peak stress as the quantity to be minimised in a stress based objective function. However our experience indicates that the above

objective function almost always achieves better convergence, avoiding local minima. The key feature of this choice of objective function is that it takes advantage of the constant stress characteristic of optimal geometries where achievement of uniform boundary stress is known to be synonymous with minimising the peak stress value [10,11, 1517]. Eectively, this objective function mimics gradientless shape optimisation approaches, which are based on an analogy with biological growth/wastage [10,11,17]. The version of the objective function given above uses responses in both load transfer regions simultaneously, as indicated by the rst and second summations. It is interesting to note that for typical sectional stress values across the adhesive thickness and away from corner singularities, the variation is less than 10% with the centre row having the highest values. This applies for the thick adhesive regions of the solution shapes as well as for the initial congurations. As can be expected, the objective function is somewhat sensitive to the element mesh that has been chosen in that its absolute value depends on the number of terms, and a ne mesh leads to a higher numerical value. In minimising the objective function its absolute value (particularly for early iterations) is not important and regardless of its absolute value, the ideal minimum targeted is zero. To achieve satisfactory optimisation solutions certain controlling parameters were specied for the MSC/ NASTRAN code. These controlling parameters include the allowable change in the design variables for each iteration. Modication of this parameter eectively controls the convergence rate and also the possibility of mesh distortion. A larger value will generally result in less run time, however, the solution runs the risk of distorting the mesh due to the larger allowable shape change. This parameter was set so as to limit the maximum outer boundary movements, between iterations, to about 1.5 mm. It should be noted that there are two measures of convergence, both of which are important to the process. The change in objective function between iterations is the measure used for terminating the process. The other measure is the absolute value of the objective function at conclusion. Ideally this value should be zero and all the stress terms equal. In practice the process always converges to the extent that no further progress can be made when two iterations give the same objective function result. At this conclusion however there can hypothetically be any amount of nonuniformity in the target stress terms. If the problem is badly dened or the target stresses not sensitive to the basis shapes the process may perform only one iteration with no improvement in objective function and therefore stop without changing the shape. Hence clearly the key check on the quality of the solution, is the degree of

R.H. Kaye, M. Heller / International Journal of Adhesion & Adhesives 22 (2002) 721

11

uniformity of the stresses, not wether or not the solution process has stopped. It worth noting that use of the term least-squares to describe the objective function can potentially have a misleading connotation in the context of the current work, as this term more commonly associated with data tting and determination of an approximate line of best t. However, this is not the case in the present work, as use of this function does not itself introduce estimation. Here the intent and implementation of this particular objective function is to reduce the stress deviation from a constant, to a value close to zero, by forcing the stresses to be more uniform. As discussed previously, subject to certain conditions, achieving stress uniformity indicates that a true optimal has been reached. 2.3.2. Design variables and basis shape vectors A key feature of the optimisation process is the way that shape changes are dened as design variables. For all but the simplest of cases, it is not practicable to have one design variable per boundary node. This would result in an excessive number of design variables, most of which would have zero sensitivity with respect to any suitable objective function. Instead, a set of displacement elds was generated by using an auxiliary model with dummy loads applied normal to the movable boundary. These displacement elds are used to link a number of nodal movements to a single design variable. The need to link the nodal movements in this way introduces a further user interaction to the optimisation process. Here the user must choose the number and nature of displacement elds so as to give the optimiser enough shape variability to achieve the optimal solution. It would be conceptually clearer if the optimiser could vary the nodal positions without restriction to arrive at the best shape, but the large number of nodes and the nature of the numerical process preclude this. The displacement elds are determined by an auxiliary model method (with some trial and error) such that the solution achieves the constant stress characteristic discussed above.

The auxiliary model must have the same geometry and node numbers as the primary model however, the material and element properties can be varied so as to give suitable shapes. It is run once only using the standard linear elastic option and the deformation output is stored and used by the optimisation process as required. It should be noted that each load is applied individually as a separate load case and this induces a displacement eld near the location of the applied load. These displacement elds are called basis shape vectors in NASTRAN and can be considered as a set of vectors Tj as follows: 2 3 DG1x ; DG1y 6 :::::::::::::::: 7 6 7 2 Tj 6 7; 4 :::::::::::::::: 5 DGKx ; DGKy where j is the basis shape vector number, DGx and DGy are the nodal displacements in the x and y direction, respectively, and K the total number of nodes (same in both models). Fig. 2 shows a typical displaced mesh due to the application of one dummy loadcase to the auxiliary model. Conveniently, the nature of elastic deformation has led movement of the non-boundary nodes, which is essential to avoid internal mesh distortion. The remainder of the model is unloaded. Typically, for the analysis cases, 25 loadcases were used to generate 25 displacement elds and consequently 25 design variables were dened in the optimisation data le. These load application points, along the movable part of the boundary or interface, are shown in Fig. 3. While the auxiliary model provides absolute displacements, which are dependent on the size of the applied dummy load, the whole displacement eld is scaled up or down as required by the optimisation process. At each iteration i 1; the new shape comes from taking the nodal positions at iteration i; and adding the displacements from all of the basis shape vectors multiplied by their scaling factors, as given by Eq. (3)

Fig. 2. Typical basis shape in joint region.

12

R.H. Kaye, M. Heller / International Journal of Adhesion & Adhesives 22 (2002) 721

Fig. 3. Finite element mesh and loading arrangement for generating basis shape vectors in auxiliary model: (a) taper region, (b) crack region, and (c) taper region alternative solution in 8(c).

below 2

7 6 7 j1 6 6 ::::::::::::: 7 6 ::::::::::: 7 X 76 7 6 Xj Tj ; 6 ::::::::::::: 7 6 ::::::::::: 7 5 4 5 jJ 4 GiKx ; GiKy Gi1 ; Gi1 Kx Ky where 2 3

Gi1 ; G1y 1x

i1 3

Gi1x ; Gi1y

3 3

6 7 B6 ::::::::::: 7C 6 ::::::::: 7 B6 7C 6 7 6 ::::::::: 7 gB6 ::::::::::: 7C: @4 5A 4 5 i i1 df =dXJ XJ Here Gx and Gy are the x and y coordinates, respectively, of nodes k 12K (all nodes in model), i the iteration number, j the design variable number in the range j 12J; X is a design variable (i.e. multiplier applied to T), df =dX is the sensitivity of the objective function with respect to a design variable, and g is a function to represent the search algorithm described in Ref. [14].

i1 X1

02

i df =dX1

31

The scaling factors Xj (one per basis shape) are the unknown design variables to be found by the optimisation process. As the scaling factors relate to signicant shape changes rather than individual nodal movements their sensitivities are less likely to be zero. As it is only the nodal displacements that are used from the auxiliary model, the elements and their properties do not need to be the same as for the structural model. Hence these can be selected so as to give a smooth deformation over a suitable region. In the cases presented, circular beam elements have been added along the movable boundaries of the auxiliary model to avoid shape discontinuities caused by the point loads. The stiness of the beam elements in relation to the neighbouring 2D elements was chosen to give a suitable shape and region of deection as shown in Fig. 2. If the beam elements are too sti relative to the plate elements, the resulting displacement elds are not suciently localised. Hence circular beam elements of 25 mm radius were used in conjunction with plate elements of 0.5 mm thickness, noting that both had the same elastic modulus. Here it was necessary to reduce the plate element thickness rather than increase the

R.H. Kaye, M. Heller / International Journal of Adhesion & Adhesives 22 (2002) 721

13

beam element radius in order to avoid extremely low aspect ratios of the beam elements. Such extreme proportions (i.e. diameter/length >1000 approximately) were found to result in erroneous bending deections having a slope discontinuity at the load point. The beam elements were typically 0.25 mm in length and the dummy load points were typically 3 mm apart. It should be noted that there is no requirement that the nodes used as application points for dummy loads need to be related in any way to those used for evaluation of the objective function. Also, 3 and 4 noded 2D elements in the structural and auxiliary model were not up-graded to 6 and 8 noded elements, as 3 noded beam elements were not available in the FE code used. Clearly, an important requirement of the numerical approach is the appropriate choice of the basis shapes and particular care must be taken. However, if a poor choice had been made, it would have been evident in the nal stress distribution (i.e. non-uniformities with locations dependent on auxiliary model load application points). As a guide 1015 basis shapes with centres equispaced along the movable boundary is a good starting point. 2.3.3. Constraints Certain constraints are dependent on the particular conguration case and are discussed in the following sections. However, a constraint on minimum adhesive thickness of 0.15 mm was consistent across all analyses.

predicted adhesive stress peak is 25.8 MPa as compared to 24.0 MPa given in Ref. [5]. Also importantly for the present context, the opening displacement of the centreline node, half way through the plate thickness in the x direction was 0.0086 mm (equivalent to the total opening of 0.0172 mm in the full conguration). For the idealisation of 2D crack patching, the stress intensity factor in the inner adherend is directly related to the opening displacement by the following equation from [2] q KT 1=2Ep fsD; 4 whereKT is an upper bound stress intensity factor for a long crack under a patch, Ep the elastic modulus of patch, f the stress reduction in the uncracked plate due to the patch (factor less than 1) at the prospective crack location, s the remote applied stress, and D the crack opening due to the remote applied stress. 3.2. Optimisation analysis In this case all 25 design variables were active (but restrained from going negative). For the taper region the doubler was allowed to vary in thickness, while the adhesive thickness was kept constant at 0.15 mm. This choice was made to allow direct comparison with some existing published solutions. However in the crack region, both the doubler and the adhesive thickness were allowed to vary. A constraint on crack opening was applied at the centreline node, half way through the plate thickness, to maintain the same repair eectiveness as the nominal repair. Hence the constraint applied was D=2 o0:0086 mm; 5 where D=2 is the displacement in the x direction at the node described above. The overall solution shape obtained after 7 iterations for the total geometry is shown in Fig. 4(c). Figs. 4(d) and (e) show more detail for the taper and crack/joint regions, respectively. The resultant von Mises adhesive stress results are given in Fig. 5. For the taper region the shape obtained is very similar to the 2D FE solutions given in other investigations [10,11]. Here there is a rapid reduction in doubler thickness at the start of the taper leading to a very ne tip at the end of the doubler. In the taper region the process has arrived at a good result, in reducing the peak stress and rendering the stress distribution relatively uniform. It can be seen that the initial peak stress of 14.4 MPa has been reduced by about 50%. The small deviations from complete uniformity in adhesive stresses in the taper region are mainly due to the fact that in this case, where only the doubler thickness is allowed to vary, there are low sensitivity values between the doubler shape and the adhesive stresses as the optimal solution is approached. If the taper region was increased in length the adhesive

3. Symmetric crack repair with aluminium doublers: case 1 In case 1 the geometry representative of a doublesided crack repair was modelled, where both the doublers (outer adherends) and the plate (inner adherend) are aluminium alloy (Fig. 4(a)). The aim was to optimise the repair shape to minimise peak von Mises adhesive stresses, such that the repair eectiveness was unchanged as compared to a nominal geometry. 3.1. Analysis of initial geometry For this problem the initial thickness of the doubler has been selected to be equivalent to that of the plate. The nite element mesh used in the quarter symmetric analysis is shown in Fig. 4(b). A very dense mesh renement was used to obtain accurate stress results, as well as represent small geometric changes in the subsequent optimisation analyses. The stresses obtained from this analysis are presented in Fig. 5, and are as expected. For the taper region they agree with those presented in Ref. [10,11] where the nite element code PAFEC was used. For the joint region they also agree well with other published FE results. For example the

14

R.H. Kaye, M. Heller / International Journal of Adhesion & Adhesives 22 (2002) 721

Fig. 4. Conguration and nite element mesh for case 1, symmetric crack repair with aluminium doublers: (a) conguration, (b) initial shape (quarter region), (c) optimised shape (quarter region), (d) optimised shape in taper region, and (e) optimised shape in joint/crack region.

stresses would have been further reduced as the same load transfer would take place over a longer region. The solution stress plot given in Fig. 5 also shows a good result in the crack/joint region with a near constant von Mises stress in the adhesive. A very large stress reduction of about 70% is evident in this linear elastic analysis (in practice, for this load level, the stress peak for the initial shape would be smaller due to local yielding of the adhesive). While the von Mises stress has

been used in the objective function, the resulting shapes are more easily understood when it is noted that the shear component dominates in the initial conguration, and the reductions in von Mises stress come about mainly by reduction in the shear component. The results in the Table 1 demonstrate this point, where the shear and peel stress magnitudes are given, near the ends of the two load transfer regions, respectively.

R.H. Kaye, M. Heller / International Journal of Adhesion & Adhesives 22 (2002) 721

15

Fig. 5. Von Mises stress distribution in adhesive for symmetric crack repair with aluminium doublers (case 1).

Table 1 Selected adhesive peel and shear stress components for case 1 in taper and crack/joint regions Conguration Stress in taper region at x 0:1 mm Shear stress (MPa) Initial shape Optimised shape 8.1 4.2 Peel stress (MPa) 0.8 0.3 Stress in crack/joint region at x 59:5 mm Shear stress (MPa) 14.7 2.2 Peel stress (MPa) 2.4 3.2

For the joint region there are some interesting features of the optimal shape. Firstly, the increase in adhesive thickness is as expected due to the high adhesive shear stress near the centreline, since typically an increase in local adhesive thickness will relieve adhesive shear stresses. Secondly, the growth in doubler thickness is a less obvious result as it has occurred not merely to help satisfy the crack opening constraint but to reduce adhesive shear stress. For comparison purposes another analyses was also undertaken with the constraint omitted, and the resulting shape was very similar with D=2 increasing by only 10%. Hence the reduction in adhesive shear stress occurs because the doubler strain in the x direction is reduced where its thickness is increased and there is locally less strain miss-match between the doubler and plate. Both these local increases in thickness have caused the load transfer through the adhesive to be diverted away from the joint symmetry line at x 60 mm, and to be distributed more evenly over the 20 mm length comprising the crack/joint region. The constraints and the problem conguration naturally bound the taper region solution. It is clear that the process has headed towards a unique optimal shape and come quite close as evidenced by the resulting stress plot. In the joint/crack region the doubler and adhesive thicknesses can continue to grow indenitely and there is no natural stopping point. In this case and cases 2 and 3 that follow the process has stopped when the optimiser can not nd further improvement, i.e. the

sensitivities have dropped to zero. The solution shape is not unique in the same way as the taper region solutions but is subject to the criterion for stopping the process. An additional constraint on overall thickness would have been another way to limit the shape changes.

4. Symmetric crack repair with boron/epoxy doublers: case 2 This conguration is the same as for case 1 except for the modelling of boron/epoxy doublers instead of the aluminium alloy doublers (Fig. 6(a)). Again the aim is to optimise the repair shape to minimise peak adhesive stresses, such that the repair eectiveness is unchanged as compared to a nominal geometry. For the analysis of the initial geometry the nite element mesh was identical to that given for the aluminium doubler case, except the doubler thickness has been reduced everywhere by a factor of three, and appropriate orthotropic material properties were used for the doubler. The stresses obtained from this analysis are presented in Fig. 7. They are as expected and are very close to the case of the aluminium doubler given in Fig. 5. For example here the peak stresses in the taper region and joint regions are 13.2 and 26.7 MPa, respectively, as compared to 14.4 and 26.2 MPa for the aluminium doubler case. For the optimisation analysis the same solution procedure as for case 1 was used. The solution shape

16

R.H. Kaye, M. Heller / International Journal of Adhesion & Adhesives 22 (2002) 721

Fig. 6. Conguration and nite element mesh for case 2, symmetric crack repair with boron/epoxy doublers: (a) conguration, (b) optimised shape (quarter region), (c) optimised shape in taper region, and (d) optimised shape in joint/crack region.

Fig. 7. Von Mises stress distribution in adhesive for symmetric crack repair with boron/epoxy doublers (case 2).

obtained after 6 iterations for the total geometry is shown in Fig. 6(b). Figs. 6(c) and (d) show more detail in the taper and joint regions, respectively. It can be seen that the optimal doubler prole is close to that of case 1 except scaled down to one third the thickness. Reduc-

tion in adhesive shear stress is again the dominant eect with the peel components being smaller than for case 1 due to the thinner doubler. Consequently, as can be expected, the adhesive stress distribution shown in Fig. 7 for both the taper and the crack regions are very close to

R.H. Kaye, M. Heller / International Journal of Adhesion & Adhesives 22 (2002) 721

17

those obtained for case 1. Here, for example, the peak stresses for the optimal solution for the taper region and joint regions are 7.8 and 9.5 MPa, respectively, slightly higher than 6.8 and 8.9 MPa for the aluminium doubler case. In the present analysis the doublers have been modelled using idealised continuous orthotropic boron/ epoxy. In practice the doublers are multilayer laminates, where each layer is about 0.13 mm thick. In both load transfer regions, for the optimal solution, the thickness changes are equivalent to ve layers, so there is some scope for manufacturing a laminate that approximates

the idealised solution shape. Obviously the thicker an optimal patch, the more closely it can be approximated by a multilayer laminate.

5. Non-symmetric crack repair with a boron/epoxy doubler: case 3 In this case we modelled the conguration where the doubler is bonded to one side of the plate only (Fig. 8(a)). Application of the remote stress will result

Fig. 8. Conguration and nite element mesh for case 3, non-symmetric crack repair with a boron/epoxy doubler: (a) conguration, (b) optimised shape (half region), (c) optimised shape in taper region, (d) optimised shape in joint/crack region, and (e) alternative optimised shape in taper region, case 3a

18

R.H. Kaye, M. Heller / International Journal of Adhesion & Adhesives 22 (2002) 721

in secondary bending, and hence a more severe adhesive stress distribution, (particularly for peel) will result. Again the aim is to optimise the repair shape to minimise peak adhesive stresses, such that the repair eectiveness is unchanged as compared to the nominal geometry. 5.1. Analysis of initial geometry For the analysis of the initial geometry the nite element mesh is identical to that given for the preceding case 2, except that the through-thickness symmetry restraint, along the line y 0 has been removed. Hence there is no support to eliminate the out of plane bending induced by the presence of the single doubler. Also, at the stress application region, out of plane rotation has been restrained to represent a plate, which is globally at but able to deect near the crack/joint region. Hence the thickness of the inner adherand is now 2 mm, while the doubler thickness and remote stress remain unchanged at 0.67 and 100 MPa, respectively. The von Mises stresses obtained from this analysis are presented in Fig. 9. It can be seen that in the taper region the stresses are higher as compared to the previous cases. For example the peak here is 30.0 MPa as compared to 14.4 MPa previously. Also in the joint region the adhesive stresses are higher, with a peak of 64.0 MPa as compared 26.2 MPa previously. The opening displacement of the centreline node half way through the

plate thickness in the x direction was 0.096 mm, as compared to 0.0086 mm previously. 5.2. Optimisation analysis For the optimisation analysis the same solution procedure as for case 2 was used except that basis shapes were modied to allow non-zero movable boundary slopes at the centreline. This was necessary because of the very large initial stress concentration, however it produces, as expected, a slope change over the crack. The solution shape obtained after 7 iterations for the total geometry is shown in Fig. 8, while the von Mises stresses are given in Fig. 9. In the taper region the optimal doubler prole is similar to that of case 1, although the adhesive stress range is 2.611.9 MPa, which is marginally higher than that of the previous cases due to the eects of bending. In the joint region the optimisation has produced a radical change of shape due to the very high initial adhesive stresses. The adhesive and doubler thicknesses have increased by a factor of three. For the joint region the optimisation process has provided stress reductions, which are larger than for the other examples. Here the stress reduction is approximately 80%. The results given in Table 2 highlight the extent to which the presence of bending has changed the adhesive stresses in both load transfer regions. All values are higher and while the shear stress component is still dominant for the initial shape in the

Fig. 9. Von Mises stress distribution in adhesive for non-symmetric crack repair with boron/epoxy doubler (case 3 and 3a).

Table 2 Selected adhesive peel and shear stress components for case 3 in taper and crack/joint regions Conguration Stress in taper region at x 0:1 mm Shear stress (MPa) Initial shape Optimised shape 16.5 6.5 Peel stress (MPa) 3.1 1.9 Stress in crack/joint region at x 59:5 mm Shear stress (MPa) 38.3 9.4 Peel stress (MPa) 40.4 15.6

R.H. Kaye, M. Heller / International Journal of Adhesion & Adhesives 22 (2002) 721

19

taper region, shear and peel have similar magnitudes in the crack region. The optimisation process has reduced both components by similar ratios in both regions. 5.3. Alternative taper region solution (case 3a) In the preceding case, the large variation in adhesive stresses in the taper region is due to this solution having an excessively long ne taper with thickness less than 0.01 mm for much of its length. It is clear that if desired,

with an increased number of basis shapes, a more uniform stress solution could be determined. However, a single layer of boron/epoxy is always about 0.13 mm thick, so instead an analysis for a more practical solution was performed for the taper region only. This conguration maintains a minimum 1 layer boron/ epoxy thickness and allows variation in adhesive thickness (unlike previous cases) to spread the load transfer. The minimum limit on the boron thickness dictates a shorter taper load transfer length of 6 mm.

Fig. 10. Conguration and nite element mesh for case 4, symmetric lap-joint with at outer boundary: (a) conguration, (b) optimised shape (quarter region), (c) optimised shape in taper region, (d) optimised shape in joint/crack region, and (e) optimised shape in joint/crack region for improved solution case 4a.

20

R.H. Kaye, M. Heller / International Journal of Adhesion & Adhesives 22 (2002) 721

The solution shape given in Fig. 8(e) was achieved using the initial geometry and auxiliary model from Fig. 3(c) and solving for the taper region on its own. The minimum boron thickness was maintained by constraining the design variables to non-negative values. The evaluation region for the objective function was reduced so as to only include the shorter 6 mm length taper load transfer region. A triangular adhesive llet was used at the end of the doubler so that the adhesive stress nearby was not inuenced by a large stress concentration that would otherwise occur at the end of the adhesive. The stress plot in Fig. 9 shows this more practical boron/ epoxy repair design gives better convergence than the standard case 3, but as expected, at a higher average value of 14.5 MPa. However there is still is a large adhesive stress reduction as compared to the initial geometry.

case 1, except for the following changes; (i) the outer doubler surface is kept at by constraining design variables 1016 to zero, (ii) the constraint on joint opening is removed (Fig. 10(a)). As before, basis shapes with zero centreline slope were used, and the initial geometry and material are the same as in case 1. The resultant optimal shape obtained after 5 iterations is given in Fig. 10, while the stresses are given in Fig. 11. As was expected the solution shape and stresses for the taper region are virtually the same as obtained in case 1. In the joint region, the stress concentration has been reduced by 48%, however the restriction on the doubler outer boundary has reduced the convergence of the adhesive stresses and a signicant concentration is still evident over the joint. Hence it is desirable to improve the quality of the solution. 6.2. An improved solution (case 4a)

6. Symmetric lap-joint analysis with constrained shape of aluminium doubler : case 4 6.1. Initial optimal solution In this case it has been dened to represent a joint rather than a crack repair. The problem is the same as

Two strategies were used to improve the solution, particularly for the joint region. The rst was to remove terms from the objective function that used stresses from elements which could not be brought to the average value. The other strategy was to further reduce the number of terms by only using element stress values at some wider spacing of position along the bond line

Fig. 11. Von Mises stress distribution in adhesive for symmetric lap-joint with constrained aluminium doubler shape (case 4) and improved solution (case 4a).

Table 3 Positions x along bond line of element centroids used for evaluating objective function in case 4a Position x in taper region (mm) Position x in crack/joint region (mm) 0.1 4.9 50.5 55.5 0.2 5.9 51.3 56.0 0.4 7.0 52.0 56.5 1.0 8.3 52.8 57.5 1.6 9.7 53.5 58.0 2.3 11.2 54.0 58.5 3.1 12.9 54.5 59.0 3.9 14.9 55.0 59.3

R.H. Kaye, M. Heller / International Journal of Adhesion & Adhesives 22 (2002) 721

21

instead of using all consecutive elements. A quantity of 16 terms in the objective function, for each of the two load transfer regions, was found to be the most successful. Distances from the end of the doubler for stress points used by the objective function are given in the Table 3 below. Typically every 10th element centroid was used in the taper region and every 3rd in the joint region. By using these strategies and allowing for non-zero centreline slope, the preceding solution was improved and is denoted case 4a. The resultant shape and stresses were largely unchanged in the taper region. However, in the crack/joint region a very dierent shape was produced as shown in Fig. 10(e). The corresponding stresses are given in Fig. 11, where it can be seen that they are much better converged over a slightly shorter region. The stress concentration near the centreline has been eliminated.

comprising multiple discrete thickness layers are addressed. The signicant reductions in peak adhesive stresses indicated in the present work suggests that through the use of optimisation procedures, wider application of bonded repairs is possible, with realistic potential for improved structural integrity.

References
[1] Baker AA, Jones R, editors. Bonded repair of aircraft structures. Dordrecht: Martinus Nijho, 1988. [2] Rose LFR. Theoretical analysis of crack patching. In: Baker AA, Jones R, editors. Chapter 5 in Bonded repair of aircraft structures. Dordrecht: Martinus Nijho, 1988. [3] Hart-smith LJ. Adhesive bonded double lap joints. NASA CR112235, 1973. [4] Tran-Cong T, Heller M. Reduction in adhesive shear strains at the ends of bonded reinforcements. DSTO-RR-0115, Defence Science and Technology Organisation, Australian Department of Defence, September 1997. [5] Shah LP, Heller M, Wang CH, Williams JF. Reduction of plate stress concentration factors due to bonded reinforcements. Int Aerosp Conf 1997;2:66776. [6] Chester RJ, Walker KF, Chalkley PD. Adhesively bonded repairs to primary aircraft structure. Int J Adhesion Adhesives 1999;19: 18. [7] Heller M, Kaye R, Whitehead S, Lubac J. Design and stress analysis. In: Chester R, editior. Life extension of F/A-18 inboard aileron hinges by shape optimisation and composite reinforcement. DSTO-TR-0699, Defence Science and Technology Organisation, Australian Department of Defence, 1999 (Chapter 3). [8] Adams RD, Comyn J, Wake WC. Structural adhesive joints in engineering, 2nd edition. London: Chapman & Hall, 1997. [9] Ojalvo IU. Optimisation of bonded joints. AIAA J 1985;23(10):157882. [10] Kaye R, Heller M. Structural shape optimisation by iterative nite element solution. DSTO-RR-0105, Defence Science and Technology Organisation, Australian Department of Defence, June 1997. [11] Heller M, Kaye R, Rose LRF. A gradientless nite element procedure for shape optimisation. J Strain Anal 1999;34(10):323 36. [12] Groth HL, Norlund P. Shape optimisation of bonded joints. Int J Adhesion Adhesives 1991;11(4):20412. [13] Kaye R, Heller M. Investigation of shape optimisation for the design of life extension options for an F/A-18 470 airframe FS bulkhead. J Strain Anal 2000;35(6):493505. [14] Moore GJ. Design sensitivity and optimisation. The MacNealSchwendler Corporation, 1994. [15] Heywood RB. Photoelasticity for designers. Oxford: Pergamon Press, 1969 (Chapter 11: Improvement of Designs). [16] Schnack E. An optimisation procedure for stress concentrations by the nite element technique. Int J Numer Meth Eng 1979;14:11524. [17] Mattheck C, Burkhardt S. A new method of structural shape optimisation based on biological growth. Int J Fatigue 1990;12(3):18590.

7. Conclusions In the present work, an automated sensitivity-based shape optimisation procedure has been developed for the optimal design of free-form bonded repairs and lapjoints, with the aim of achieving reduced adhesive stresses. The approach has been demonstrated through application to a number of single and double-sided congurations where both the shapes of the adhesive layer and the outer adherand are allowed to vary. Signicant improvements over conventional designs are obtained, as assessed by the reduction in peak adhesive stresses. For many real cases these reductions are sucient to keep adhesive stresses in the elastic range which would make fatigue failures unlikely. The taper region solutions all show the benet of a very ne taper at the end of the doubler that has the eect of spreading the load transfer more evenly. In the crack/joint region all cases examined indicated adhesive stress reductions through increased doubler and adhesive thicknesses. Here the increase in doubler thickness reduces adhesive shear stress due to the locally reduced extension under load, while the increase in adhesive thickness reduces adhesive shear stress by providing local shear compliance and shedding the shear load away from the concentration. The focus has been primarily on investigating continuously variable shapes, hence it is envisaged that in future work, optimal solutions will be developed where the constraints imposed by having a doubler

You might also like