You are on page 1of 18

Journal of Adhesion Science and Technology 24 (2010) 19591976

brill.nl/jast

Robust Shape Optimization of Tubular Butt Joints for Characterizing Thin Adhesive Layers under Uniform Normal and Shear Stresses
D. Castagnetti , A. Spaggiari and E. Dragoni
Department of Engineering Sciences and Methods, University of Modena and Reggio Emilia, Via Amendola 2 Campus S. Lazzaro Pad. Morselli, 42100 Reggio Emilia (RE), Italy

Abstract Thin-walled tubular joints, bonded end to end, are commonly used specimens to measure the mechanical properties of thin adhesive layers subjected to uniform shear stress distributions. Unfortunately, the application of an axial loading to this geometry leads to strong stress concentrations at the edges of the adherendadhesive interface. This drawback undermines the use of this test for characterizing adhesives under biaxial stress conditions. With the aim of removing these stress concentrations, this paper suggests the introduction of stress relieving grooves on the internal and external surfaces of the tubular adherends. The optimal shape of the groove is identied following the Taguchi robust optimization technique. Via nite element analyses, the stress concentrations at the edges of the adherendadhesive interface are calculated. Many geometries are examined for different adherend and adhesive properties (noise factors) in order to identify the groove shape that minimizes the stress concentrations for all experimental conditions. The analysis shows that a shallow V-shaped groove close to the adherendadhesive interface smoothes signicantly the stress peaks due to axial loading. With this simple modication, a tubular butt joint becomes a universal specimen for applying any combinations of reasonably uniform shear and normal stresses to thin adhesive layers. Koninklijke Brill NV, Leiden, 2010 Keywords Tubular butt joint, stress concentrations, robust design, shape optimization

1. Introduction The mechanical properties of adhesives, which are used as an input for the various stress prediction techniques and are the basis for the application of the failure criteria, have to be determined unambiguously. Controversy still exists whether adhesive properties are the same in the thin layer form as in the form of bulk specimen. Adhesives are normally used in joints in a thin layer form. However, if their mechanical properties are measured in thin layer form they may be affected by
To whom correspondence should be addressed. Tel.: 0039 0522 522634; Fax: 0039 0522 522609; e-mail: dcastagnetti@unimore.it
Koninklijke Brill NV, Leiden, 2010
DOI:10.1163/016942410X507687
*

1960

D. Castagnetti et al. Journal of Adhesion Science and Technology 24 (2010) 19591976

the transverse restraint imposed by the adherends. Moreover, when in a thin layer form, adhesives may become anisotropic or may be subjected to some chemical contamination due to the adherends and their protective coatings [1]. The bulk form eliminates the above problems but produces quite different conditions for the adhesive with respect to its normal state in joints. Moreover, it is difcult to produce many polymeric adhesives in the bulk form because of the signicant increase of temperature which occurs during polymerization. The temperature increase originates from the exothermic reactions involved in polymerization of the adhesive. Tests on adhesives both in the thin layer and bulk specimen forms are commonly used. Measuring the mechanical properties of a thin adhesive layer subjected to uniform stress distributions provides important information for predicting the adhesive behavior under working conditions. Thin-walled tubular joints, bonded end to end and subjected to torsional loading, are commonly employed to characterize the adhesives under fairly uniform shear stress elds [1, 2]. Unfortunately, the application of an axial loading to the same geometry leads to a distribution of normal stresses in the adhesive with strong concentrations at the edges of the bondline [2, 3]. The parameters that affect these stress concentrations, which arise at the joint interface between dissimilar materials, have been studied in the literature. The reports that can be retrieved deal both with the general case of a dissimilar materials joint [48] as well as with specically bonded joints [913]. Theoretical analyses performed in [48] show that the stress concentration depends signicantly not only on the elastic properties of the materials but also on the local geometry of the joint interface. More specically, in [68] it is shown that a convex joint interface is quite effective in eliminating the free edge stress singularities. In particular, Wang and Xu [8] show that by changing the geometry from straight-edged joints to convexedged axisymmetric joints of dissimilar materials the free edge stress singularities disappear. Similar investigations were performed also on bonded joints where a singular stress eld originated at the adherendadhesive interface due to the geometry and to the elastic mismatch of the components. Reedy and Guess [9] studied the effect of bond thickness on tensile strength for a butt joint with two stainless steel rods. They proposed a fracture criterion based on a critical interface corner stress intensity factor. Koguchi [10] investigated the order of stress singularities at the vertex of a three-dimensional dissimilar materials joint. Penado [11] evaluated the singular stress intensity factors in bonded joints with and without adhesive llets. Kotousov [12] studied the effect of a thin plastic adhesive layer on the singular stress eld at the interface corner of an adhesively bonded bi-material wedge. Leguillon et al. [13] examined the failure initiation in a bonded joint between two steel plates. The initiation of failure was assessed by estimating the critical values of the stress intensity factor of the elastic singular stress elds appearing in certain regions of the structures.

D. Castagnetti et al. Journal of Adhesion Science and Technology 24 (2010) 19591976

1961

Stress concentrations which occur in bonded joints are particularly misleading when the adhesive response is studied under a wide range of loading conditions. In order to overcome this problem many test xtures aimed at providing a uniform stress eld have been proposed in the literature. Fischer and Pasquier [1] investigated the adhesive behavior by performing torsion tests on tubular bonded joints under different test temperatures. They found that the shear moduli of the bulk adhesives were comparable with those measured in the bondline. Cognard and coworkers [14, 15] developed a modied Arcan xture which enables compression or tension to be combined with shear loads and allows to characterize thin adhesive layers up to failure. In [16], Cognard presents a modication in the geometry of the substrates of the modied Arcan test in order to lower the stress in the adhesive near the free edge of the joint. This modication consists in a beak, machined on the adherend at the interface with the adhesive, with an angle in the range between 30 and 45 . A comparison between the modied Arcan xture and the standard Thick Adherend Shear Test (TAST) conguration is provided in [17]. The results presented in [17] indicate differences in the measured shear behaviours. In particular, the TAST specimen provides a nonlinear behaviour which is different from the one of the modied Arcan xture because the damage initiates near the adhesive/adherend interface rather than within the adhesive. Two limitations can be highlighted in the modied Arcan xture. The rst concerns the fact that the beaks introduced in the specimen cancel out the stresses along the perimeter mainly in the mid-plane of the adhesive but much less effectively at the adherendadhesive interface. The second is the complex and expensive manufacturing of the beaks on the specimen. In order to overcome these limitations, this paper examines a new geometry of the tubular butt joint. The aim is to produce in the thin adhesive layer a distribution of normal stresses under axial loading as uniform as the shear stresses under torsional loading. The stress uniformity must be achieved at the free edge not only in the middle of the adhesive layer but also at the adherendadhesive interface. This goal is accomplished by introducing grooves on both the internal and external surfaces of the adherends so as to increase the local compliance of the adherends close to the stress concentrations in the adhesive. The work follows Taguchis robust design method [1821] to obtain a single optimal groove that gives uniform adhesive stresses irrespective of the experimental conditions (size of the specimen, thickness and modulus of the adhesive, etc.). The main step in this approach is a computational analysis plan which combines efciently the geometrical parameters of the groove and the noise factors. In this main step there are four control factors varied over three levels, and three are the chosen noise factors varied over two levels. Thus eight joint congurations are examined over four different noise levels. In a nal step, the results of the main plan are analyzed to identify and validate the optimal groove producing the most robust stress uniformity in the adhesive. The process shows that a V-shaped groove, 0.1 mm from

1962

D. Castagnetti et al. Journal of Adhesion Science and Technology 24 (2010) 19591976

the adherendadhesive interface, having a depth of 1 mm and an opening angle of 30 represents the optimum shape. This shape regulates the stress at the interface of the free edge, independently of the wall thickness of the tubular adherends and of the other experimental conditions. The tubular butt joint, equipped with the proposed groove, becomes a universal specimen for applying any combinations of shear and normal stresses to thin adhesive layers. Both the internal and the external grooves can be easily manufactured on a lathe using a pre-shaped tool and turning the single adherends separately before bonding. 2. Materials and Methods Figure 1 provides a qualitative description of the structure of the tubular bonded joint under scrutiny including a three-dimensional section. The peculiarity of this joint structure is the introduction of V-shaped grooves both on the inner and outer side of each tubular adherend near the adhesive layer. This modication of the geometry was inspired by the convex-edged axisymmetric joints proposed in [8]. This groove should preclude stress singularities at the adherendadhesive interface, typical of straight-edged joints, which arise when an axial load is applied to the joint. As in the grooveless joint, the uniformity of the shear stresses is guaranteed by making the wall thickness of the tubes sufciently small with respect to the mean radius.

Figure 1. Three-dimensional cross-sectional view of the tubular bonded joint (left) with details of the stress-relieving groove (right).

D. Castagnetti et al. Journal of Adhesion Science and Technology 24 (2010) 19591976

1963

A general V-shaped groove was chosen in accordance with the shape proposed for the beaks in the modied Arcan test [1417]. The Taguchis robust design method [1821] was applied to identify the most robust shape of the groove. This shape is the one that minimizes the edge effect under axial loading of the joint, by being less sensitive to the noise factors. The robust design exploited in this study involves six steps. Starting from the denition of the system response, calculated with nite element models, the method denes the control and the noise factors. Then, the main experiment was carried out by performing a wide test campaign. The analysis of the results highlights how the control factors affect the response and provides indications about the optimal congurations which were analyzed in the last verication experiment. 2.1. System Response The primary response considered as output for the robust design was dened as a stress ratio. This ratio was calculated between the peak von Mises stress arising in the adhesive at the free edge of the adherendadhesive interface (point A in Fig. 2) and the axial stress remotely applied to the joint ( in Fig. 2). Henceforth, this ratio is called normalized edge stress. Since the analysis was performed in the elastic eld and failure of the adhesive was not investigated in this work the von Mises stress was used. This equivalent stress criterion was chosen because it combines the normal and the shear stress components which originate due to the axial load. The shear stress arises principally at the free edge of the adherendadhesive interface due to the Poissons effect. The secondary response of the system was the maximum normal stress on the mid-plane of the adhesive layer at its free edge (point B in Fig. 2) calculated

Figure 2. Sketch of half model of the axisymmetric bonded joint with dimensions of the joint and of the groove. The hatched area highlights the region described with the nite element model.

1964

D. Castagnetti et al. Journal of Adhesion Science and Technology 24 (2010) 19591976

and normalized over the loading stress . This secondary response was taken into account only as a verication parameter because the stress concentration on the mid-plane is much less severe than at the interface. The maximum stresses used in the response functions were calculated by means of nite element analyses. Figure 2 shows a longitudinal section view of the tubular bonded joint with the detail of the general shape of the groove analyzed. A plain strain condition was assumed in the computational model. Due to the double symmetry of the section the assumption of a plain strain condition allows the computational model to reproduce only half of the radial section (hatched area in Fig. 2) with respect to an equivalent axisymmetric model. Moreover, the plain strain condition is as accurate as the axisymmetric condition to describe the mechanical behaviour of thin-walled tubes loaded axially. This geometric consideration will be discussed in more detail in Section 2.3. Both the adherend and the adhesive are described by means of quadrilateral, quadratic, full integration plain strain elements. From a preliminary analysis two critical square areas (with a side length of 0.025 mm) were identied across the adherendadhesive interface near the free edge of the joint (point A in Fig. 2). The side length of the elements (e) meshing these two square areas was set equal to 0.125 m for a total of 200 elements along each side of the square areas. This element dimension can be assumed as a good trade-off between accuracy and computational efciency, in order to properly describe the stress concentrations that arise in practical congurations. As said at the beginning of Section 2, the stress concentrations are expected to be singular only for the reference grooveless joint considered in the main experiment (Section 2.4). On the contrary, the V-shaped groove introduced in the modied geometry of the tubular joint near the adhesive adherend interface is expected to remove such stress singularities [8]. By keeping a constant mesh pattern for all the joint congurations analyzed it is possible to provide comparable and reliable information in terms of elastic stresses. The constitutive behaviour was described as linear elastic behaviour both for the adherend and the adhesive. Symmetry constraints were imposed on the symmetry lines and a uniform normal stress with a unit value was applied on the upper side of the adherend ( in Fig. 2). The whole set of nite element analyses was implemented using the FE software ABAQUS 6.8 [22]. 2.2. Control Factors The control factors of the design are the four geometric parameters that describe the shape of the groove (Fig. 2): s the distance of the groove from the adherendadhesive interface; t the groove depth; r the curvature radius at the tip of the groove; the opening angle of the V-shaped groove.

D. Castagnetti et al. Journal of Adhesion Science and Technology 24 (2010) 19591976 Table 1. Levels of the control factors of Section 2.2 dening the shape of the groove s (mm) t (mm) r (mm) ( ) 0.1 0.25 0 30 0 0.2 0.5 0.5 60 + 0.4 1 1 90

1965

Table 2. Levels of the noise factors dened in Section 2.3 Ea (MPa) E (MPa) sa (mm) Symmetry R (mm) w (mm) L (mm) e (m) 1500 210 000 0.1 Yes Not dened 5 10 0.125 + 3000 69 000 0.4

Three levels for each factor were considered as presented in Table 1. 2.3. Noise Factors Table 2 presents the noise factors affecting the response of the joint. These factors are: Ea the elastic Youngs modulus of the adhesive; E the elastic Youngs modulus of the adherends; sa the adhesive thickness; Symmetry the symmetry of the groove with respect to the mid-plane of the adhesive; R the average radius of the tubular adherend; w the wall thickness of the tubular adherend; L the adherend length in the longitudinal direction; e the element side length in the computational model. In order to assess which of these parameters actually inuenced the stress concentration arising at the free edge of the joint, preliminary analyses (not reported

1966

D. Castagnetti et al. Journal of Adhesion Science and Technology 24 (2010) 19591976

here for the sake of brevity) were performed. On the basis of these analyses the more signicant noise factors for the response were found to be: the elastic modulus of the adhesive Ea , the elastic modulus of the adherend E, and the adhesive layer thickness sa . The elastic Youngs moduli were combined in the ratio E/Ea in order to obtain a more comprehensive noise factor. The symmetry of the groove, with respect to the mid-plane of the adhesive, was maintained for all joint congurations so as to obtain a pair of grooves both on the top and bottom adherends. The absence of symmetry condition in fact would prevent reduction of the stress concentration at the interface on the side without the groove. The mean radius R of the tubular adherend is related to the curvature of the joint. The uniformity of torsional shear stresses is guaranteed if the radius R is signicantly higher than the adherend wall thickness w. A high R/w ratio ensures that the normal stress distribution under axial loading is negligibly affected by the value of the mean radius. These considerations allowed to disregard the mean radius of the adherend R as a noise factor. Hence, the nite element models were implemented with the assumption of plain strain condition as described in Section 2.1. A preliminary investigation performed on the reference grooveless joint showed that the plain strain and axisymmetric conditions provide exactly the same results for an R/w ratio higher than 5. This is because under the purely axial load the radial displacement of the thin wall is small and produces a circumferential strain close to zero when divided by the large mean radius of the tube [23]. Due to the very localized character of the stress concentration at the edges, it was assumed that the wall thickness w had a negligible inuence on the response. A constant value w = 5 mm, which is used also by other researchers [1], was adopted throughout the optimization process. In order to conrm this hypothesis, an analysis with a double thickness w = 10 mm was performed on the optimal groove geometry in Section 2.6. The adherend length L was set to a constant value equal to two times the width of the radial section w. It was veried that this dimension avoided mutual perturbation in the stress eld between the joint interface and the loaded upper boundary of the adherend. Also, the side element length e near the free edge of the adherendadhesive interface was kept constant in all models at the value (0.125 m) that provided reasonable mesh convergence (see Section 2.1). In accordance with Taguchis robust design approach [18], all the signicant noise variables were varied to a great extent over two levels, as shown in Table 2. The values chosen for the elastic modulus of the adherends correspond to an aluminium adherend (E = 69 000 MPa, = 0.3) and a steel adherend (E = 210 000 MPa, = 0.3). The values chosen for the elastic modulus of the adhesive, on the contrary, are intended to simulate a compliant adhesive (Ea = 1500 MPa, = 0.3) and a rigid adhesive (Ea = 3000 MPa, = 0.3). Although the chosen

D. Castagnetti et al. Journal of Adhesion Science and Technology 24 (2010) 19591976 Table 3. L9 orthogonal array of the control factors Standard conguration 1 2 3 4 5 6 7 8 9 s (mm) low low low medium medium medium high high high t (mm) low medium high low medium high low medium high r (mm) low medium high medium high low high low medium ( ) low medium high high low medium medium high low

1967

Youngs moduli of the adhesive are of the same order of magnitude, these values combined with the Youngs modulus of the adherends provide a wide variation of the noise factor (E/Ea ). The levels of the adhesive layer thickness were chosen in the range of practical interest for structural adhesives (0.1 mm and 0.4 mm). 2.4. The Main Experiment The control factors were combined with an L9 orthogonal array (Table 3) which is suggested as the most efcient orthogonal design to take into account four factors varied over three levels [18]. A full factorial plan was used, instead, to combine the two noise factors, leading to four combination cases, henceforth called noise levels. A total set of 36 computational analyses performed in this step was obtained by combining the L9 orthogonal array with the four noise levels, as shown in Tables 4 and 5. In addition to these analyses, as a benchmark for the results provided by the robust optimization, the reference tubular bonded butt joint without any groove on its surface was also studied. 2.5. Analysis of the Results In order to assess the more robust shape of the groove which minimizes the normalized edge stress at point A (Fig. 2), the signal to noise (S/N ) ratio for this response was calculated. Since the normalized edge stress was to be minimized a smaller the better design was adopted for the S/N ratio. For each joint conguration of Table 3, the S/N ratio was calculated with the following formula [19]: 1 S = 10 Log10 N n
n

yi2 ,
i=1

(1)

where yi represents the normalized edge stress at point A in Fig. 2 for each noise condition. In addition to the S/N ratio, the average normalized edge stress at

1968

D. Castagnetti et al. Journal of Adhesion Science and Technology 24 (2010) 19591976

Table 4. Results of the main experiment at the adherendadhesive interface Noise factors E/Ea sa (mm) Standard conguration Control factors s (mm) t (mm) r (mm) 0 0.5 1 0.5 1 0 1 0 0.5 ( ) 30 60 90 90 30 60 60 90 30 0.13 0.28 0.34 1.92 0.57 0.74 2.61 1.36 0.78 3.66 2.45 1.77 1.35 4.17 3.03 0.31 4.48 3.72 0.52 4.97 1.83 1.35 0.94 4.09 2.73 0.01 4.58 3.60 0.19 5.23 5.97 5.61 5.31 7.10 6.51 4.33 7.26 6.87 4.60 7.43 2.60 2.25 1.99 4.32 3.21 1.35 4.73 3.89 1.52 5.32 10.52 9.60 8.90 13.43 11.72 6.85 14.00 12.78 7.41 14.79 Noise level 1 23 0.1 2 140 0.1 3 23 0.4 4 140 0.4 Average S/N

Normalized edge stress

1 0.1 0.25 2 0.1 0.5 3 0.1 1 4 0.2 0.25 5 0.2 0.5 6 0.2 1 7 0.4 0.25 8 0.4 0.5 9 0.4 1 Reference grooveless joint

Table 5. Results of the main experiment on the mid-plane of the adhesive Noise factors E/Ea sa (mm) Standard conguration Control factors s (mm) t (mm) r (mm) 0 0.5 1 0.5 1 0 1 0 0.5 ( ) 30 60 90 90 30 60 60 90 30 0.16 0.11 0.09 0.24 0.03 0.14 0.37 0.17 0.15 0.54 0.20 0.14 0.11 0.44 0.31 0.01 0.49 0.40 0.02 0.55 0.12 0.02 0.02 0.40 0.21 0.16 0.46 0.33 0.13 0.55 0.42 0.37 0.35 0.52 0.46 0.25 0.54 0.50 0.28 0.56 0.23 0.16 0.14 0.40 0.25 0.14 0.46 0.35 0.15 0.55 11.90 13.63 14.43 7.66 10.50 15.70 6.60 8.67 15.33 5.21 Noise level 1 23 0.1 2 140 0.1 3 23 0.4 4 140 0.4 Average S/N

Normalized edge stress

1 0.1 0.25 2 0.1 0.5 3 0.1 1 4 0.2 0.25 5 0.2 0.5 6 0.2 1 7 0.4 0.25 8 0.4 0.5 9 0.4 1 Reference grooveless joint

point A in Fig. 2 was also calculated for each of the nine joint congurations examined.

D. Castagnetti et al. Journal of Adhesion Science and Technology 24 (2010) 19591976

1969

Starting with the S/N ratio for each joint conguration, the effect of the control factors on the S/N ratio was evaluated by calculating the average of the S/N ratios for each level of each control factor. This was aimed at identifying the optimal value of each control factor. The choice of the optimum values of levels of the control factors is presented in the Discussion section. Since the L9 array is not a full factorial plan, the optimal conguration may not be included in it. The S/N ratio and the effect of the control factors on this ratio were also calculated for the normalized edge stress arising at point B (Fig. 2). This was aimed at assessing the inuence of the groove on the mid-plane of the adhesive. 2.6. Verication Experiment The optimal joint congurations were identied by combining in all possible ways the optimal values of the control factors given by the previous analysis of the results. These congurations were validated with a verication experiment in order to estimate which one was the optimal. The same four levels of noise considered in the main experiment were taken into account for each conguration. The computational analysis model was implemented for each conguration exactly as in the previous step. The analysis of the results was performed in the same way as for the main experiment. From this analysis, which of these optimal congurations performed the best in terms of robust minimization of the normalized edge stress was assessed. A further verication experiment was performed in order to assess the validity of the hypothesis to neglect the noise factor corresponding to the width of the adherend wall w considered in Section 2.3. The optimal conguration identied in the above verication experiment was examined by considering the same control parameters and by setting the noise factors at the level that provided the worst primary response for that conguration. The adherend wall thickness w was set equal to 10 mm. 3. Results Tables 4 and 5 summarize the results of the main experiment. Table 4 refers to the adherendadhesive interface and Table 5 refers to the mid-plane of the adhesive. The centre columns in both tables, labelled as normalized edge stress, display the system responses including the outcome from the reference grooveless joint (bottom row of the tables). The last two columns in Tables 4 and 5 provide the average response and the S/N ratio obtained from the responses of each groove geometry. Figure 3 presents the diagrams of the effect of each control factor on the S/N ratio. In each diagram the curve marked by solid squares refers to the adherend adhesive interface while the curve marked by solid diamonds refers to the mid-plane of the adhesive. Table 6 presents the optimum values of the control factors in order to maximize the S/N ratio of the normalized edge stress in the adhesive. The values in Table 6

1970

D. Castagnetti et al. Journal of Adhesion Science and Technology 24 (2010) 19591976

Figure 3. Average effects of the control factors (normalized levels) on the S/N ratios: the distance of the groove from the adherendadhesive interface, s (a), the groove depth, t (b), the curvature radius adherendadhesive at the tip of the groove, r (c) and the opening angle of the groove, (d). adhesive mid-plane. interface, Table 6. Optimum values of the control factors from Fig. 3 s (mm) t (mm) r (mm) ( ) 0.1 1 0 30 0.5 60

were identied by picking from Fig. 3 the levels of the control factors that maximized the S/N ratio (see Discussion). Figure 4 shows the sketches of the joint congurations originated by all possible combinations of the parameters collected in Table 6. Tables 7 and 8 summarize the results of the computational analyses performed on the joint congurations of Fig. 4. The layout of these tables is the same as for Tables 5 and 6.

D. Castagnetti et al. Journal of Adhesion Science and Technology 24 (2010) 19591976

1971

Figure 4. Optimal joint congurations examined in the verication step of Section 2.6 obtained by combining the optimum values of the control factors presented in Table 6. All dimensions in mm.

Figure 5 presents the distribution of the normalized edge stress along the adherendadhesive interface (Fig. 5(a)) and along the mid-plane of the adhesive (Fig. 5(b)) for three joint congurations. For each conguration of the joint only the worst case, in terms of normalized edge stress at the adherendadhesive interface, is provided. The solid black line refers to the reference grooveless joint at the fourth noise level (E/Ea = 140, sa = 0.4 mm). The solid light gray line refers to the standard conguration 6 (Table 4) also at the fourth noise level (E/Ea = 140, sa = 0.4 mm). Finally, the dashed gray line refers to conguration O3 (Table 7) at the second noise level (E/Ea = 140, sa = 0.1 mm). A normalized edge stress at the adherendadhesive interface equal to 0.89 was provided by the verication analysis performed with the aim to assess the assump-

1972

D. Castagnetti et al. Journal of Adhesion Science and Technology 24 (2010) 19591976

Table 7. Results of the verication experiment at the adherendadhesive interface Noise factors E/Ea sa (mm) Optimized conguration O1 O2 O3 O4 Control factors s (mm) t (mm) r (mm) 0.1 0.1 0.1 0.1 1 1 1 1 0.5 0.5 0 0 ( ) 30 60 30 60 0.11 0.46 0.09 0.50 0.89 0.13 1.02 0.26 0.53 0.26 0.61 0.36 1.01 3.52 0.31 3.35 0.63 1.09 0.51 1.12 2.82 5.01 4.19 4.64 Noise level 1 23 0.1 2 140 0.1 3 23 0.4 4 140 0.4 Average S/N

Normalized edge stress

Table 8. Results of the verication experiment on the mid-plane of the adhesive Noise factors E/Ea sa (mm) Optimized conguration O1 O2 O3 O4 Control factors s (mm) t (mm) r (mm) 0.1 0.1 0.1 0.1 1 1 1 1 0.5 0.5 0 0 ( ) 30 60 30 60 0.01 0.09 0.09 0.00 0.13 0.06 0.07 0.14 0.25 0.20 0.22 0.26 0.10 0.17 0.15 0.17 0.12 0.13 0.13 0.14 16.54 17.00 16.75 15.43 Noise level 1 23 0.1 2 140 0.1 3 23 0.4 4 140 0.4 Average S/N

Normalized edge stress

tion made for the noise factor w. The verication analysis was executed on the joint conguration O3 (Fig. 4(c)) with the following noise factors: w = 10 mm, E/Ea = 140 and sa = 0.1 mm. 4. Discussion Table 4 shows that for all congurations of the groove, the normalized edge stress at the interface exhibits signicant variations as a consequence of the four different noise levels tested. Lower normalized edge stresses are obtained for the rst and the third noise levels (low E/Ea ). On the contrary, when the ratio between the elastic modulus of the adherend and the adhesive (E/Ea ) is set at the higher value (second and fourth noise levels), the normalized edge stress signicantly increases.

D. Castagnetti et al. Journal of Adhesion Science and Technology 24 (2010) 19591976

1973

(a)

(b)

Figure 5. Normalized stress along the adherendadhesive interface (a) and the mid-plane of the adhesive (b) for the benchmark geometry (fourth noise level), standard conguration 6 (fourth noise level) benchmark geometry, standard conguration 6, and conguration O3 (second noise level). conguration O3.

The joint standard conguration 6 is the one that maximizes the S/N ratio calculated for the normalized edge stress at the adherendadhesive interface (primary response). Hence, it provides the most robust solution for the joint congurations included in the main experiment. Moreover, this conguration minimizes the average normalized edge stress as can be observed by the column showing the average in Table 4.

1974

D. Castagnetti et al. Journal of Adhesion Science and Technology 24 (2010) 19591976

Examination of Table 5 also shows that the secondary response (i.e., the normalized edge stress on the mid-plane of the adhesive layer) presents oscillations due to the noise levels, with the same trend as that of Table 4. However, this response, which has been considered only as a secondary verication parameter (see Section 2.1), is lower than at the interface. This is true for both the reference grooveless joint and for the joint congurations of the main experiment. Standard conguration 6 provides again the most robust solution by maximizing the S/N ratio for the secondary response. Moreover, this conguration minimizes the average normalized edge stress (Table 5). Figure 3 shows that both the primary and the secondary responses exhibit exactly the same trend in terms of the effect of the control factors, as can be noticed by the perfect parallelism of the curves. The control factors that have a higher inuence on the robustness of the solution are the groove depth, t (Fig. 3(b)), and the distance of the groove from the adherendadhesive interface, s (Fig. 3(a)). The maximum S/N ratio for the primary response (solid squares) is obtained at the lowest level for the distance s (Fig. 3(a)) and at the highest level for the groove depth t (Fig. 3(b)). The curvature radius r (Fig. 3(c)) and the opening angle of the groove (Fig. 3(d)), on the contrary, affect only slightly the primary response. For both of these, either a lower or a medium level can be chosen with a negligible inuence on the robustness of the solution. Therefore, in Table 6 the lower level value for the distance s and the higher level value for the groove depth t are provided. Both the lower and medium level values are specied instead for the curvature radius r and the opening angle . It is interesting to observe that the congurations included in Tables 7 and 8 are different from those examined in the L9 orthogonal array of the main experiment (Table 6). This is a valuable peculiarity of the Taguchi methodology, which makes it possible to discover improved solutions, previously unexplored, through the analysis of the average effects of the variables (Fig. 3 and Table 6). From Tables 7 and 8 it appears that all the optimal joint congurations (O1 O4) provide a signicant decrease of the normalized edge stress. This decrease is achieved both at the adherendadhesive interface (Table 7) and on the mid-plane of the adhesive (Table 8) for all the noise levels examined. Table 7 shows that the maximum value of the S/N ratio calculated for the normalized edge stress at the adherendadhesive interface is provided by joint conguration O3. Moreover, this conguration minimizes the average normalized edge stress. As shown in Table 8, joint conguration O3 provides also for the normalized edge stress on the mid-plane of the adhesive the highest S/N ratio and minimizes the average normalized edge stress on this mid-plane. Hence, it can be concluded that conguration O3 (Fig. 4(c)) corresponds to the groove shape that provides the most robust solution by signicantly removing the stress concentrations at the adherendadhesive interface as well as on the mid-plane of the adhesive.

D. Castagnetti et al. Journal of Adhesion Science and Technology 24 (2010) 19591976

1975

Figure 5 shows that both optimal congurations (standard 6 and O3) slightly increase the normalized stress in the centre of the adherendadhesive interface (Fig. 5(a)) with respect to the reference geometry. The decrease of the normalized edge stress, on the contrary, appears quite signicant. For standard conguration 6 this decrease is 42% while for the O3 conguration it is 82% of the value provided by the reference joint. Moreover, in the optimal conguration O3 the normalized stress is signicantly reduced both at the adherendadhesive interface as well as on the mid-plane, i.e., all along the region which is shielded by the presence of the groove. On the mid-plane of the adhesive (Fig. 5(b)), both optimal congurations slightly increase the normalized stress with respect to the reference geometry, in the central portion. The decrease in the normalized edge stress is 55% for the standard 6 conguration with respect to the reference geometry and 74% in the case of O3 conguration. In this case, however, the stress concentration is practically absent. The diagram of Fig. 5(b) highlights a drawback of the optimal conguration O3. This conguration signicantly affects the uniformity of the normalized stress on the mid-plane of the adhesive. However, as far as the measurement of the adhesive strength is concerned, the lowering of the stress on the mid-plane is far better than a nearly-singular stress concentration at the edges. The assumption to neglect from the noise factors the width of the wall of the adherend w is shown to be valid. The normalized edge stress at the adherendadhesive interface decreases only by 10% when the wall thickness w doubles. This was assessed by a verication analysis which was performed on the joint conguration O3. In this verication analysis the wall thickness was set equal to 5 mm which is twice the value considered in the main experiment (Section 2.4) and in the verication experiment (Section 2.6). In summary, the introduction of grooves having the simple geometry described in Fig. 4(c) on the internal and external surfaces of tubular adherends substantially removes the stress concentrations at the edges of the bondline. This holds true for a wide range of experimental conditions. With this modication, a tubular bonded butt joint becomes a useful and simple specimen for testing thin adhesive layers under any combinations of uniform normal and shear stresses. 5. Conclusion The mechanical properties of a thin adhesive layer can simply be measured using thin-walled tubular joints specimens, bonded end to end. A uniform shear stress distribution arises when a torsional load is applied to the joint. Unfortunately, the application of an axial loading to this geometry leads to strong stress concentrations at the edges of the adherendadhesive interface. In order to remove these stress concentrations, stress relieving grooves on the internal and external surfaces of the adherends are introduced so as to increase the local compliance of the adherends where the stress concentrations arise. The optimal shape of the groove is identied

1976

D. Castagnetti et al. Journal of Adhesion Science and Technology 24 (2010) 19591976

following the Taguchi robust optimization technique. In the main step of the method the stress concentration at the edge at the adherendadhesive interface is calculated through computational analysis. A systematic plan is implemented, which combines efciently the control parameters of the groove with the characteristic noise factors (adherend and adhesive properties). In a nal step, the results of the main plan are analyzed to identify and validate a new optimal groove producing the most robust stress uniformity in the adhesive. The analysis shows that a V-shaped groove (0.1 mm from the adherendadhesive interface, 1 mm deep and with an opening angle of 30 ) smoothes signicantly the stress peaks due to axial loading. This simple modication makes a tubular butt joint a universal specimen which makes possible to apply to thin adhesive layers any combinations of reasonably uniform shear and normal stresses. References
1. M. Fischer and M. Pasquier, Constr. Build. Mater. 3, 3134 (1989). 2. T. Sawa, in: Modeling of Adhesively Bonded Joints, L. F. M. da Silva and A. chsner (Eds), pp. 7491. Springer, Berlin (2008). 3. R. D. Adams, J. Comyn and W. C. Wake, Structural Adhesive Joints in Engineering. Elsevier, London (1984). 4. Y. Y. Yang and D. Munz, Intl J. Solids Struct. 34, 11991216 (1997). 5. Z. Q. Qian, Computers Structures 79, 5364 (2001). 6. L. R. Xu, H. Kuai and S. Sengupta, Expl. Mech. 44, 616621 (2004). 7. Z. Wu, Composite Struct. 65, 339345 (2004). 8. P. Wang and L. R. Xu, Mech. Mater. 38, 10011011 (2006). 9. E. D. Reedy and T. R. Guess, Intl J. Solids Struct. 30, 29292936 (1993). 10. H. Koguchi, Intl J. Solids Struct. 34, 461480 (1997). 11. F. E. Penado, Intl J. Fracture 105, 125 (2000). 12. A. Kotousov, Intl J. Adhesion Adhesives 27, 647652 (2007). 13. D. Leguillon, J. Laurencin and M. Dupeux, European J. Mech. A 22, 509524 (2003). 14. J. Y. Cognard, P. Davies, B. Gineste and L. Sohier, Composites Sci. Technol. 65, 359368 (2005). 15. J. Y. Cognard, P. Davies, L. Sohier and R. Creachcadec, Composite Struct. 76, 3446 (2006). 16. J. Y. Cognard, Computers Structures 86, 17041717 (2008). 17. J. Y. Cognard, R. Creachcadec, L. Sohier and P. Davies, Intl J. Adhesion Adhesives 28, 393404 (2008). 18. D. M. Bryne and S. Taguchi, Quality Prog. 20, 1926 (1987). 19. M. S. Phadke, Quality Engineering Using Robust Design. Prentice Hall, Englewood Cliff, NJ (1989). 20. G. Taguchi, Quality Reliability Eng. Intl. 12, 7374 (1996). 21. L. F. M. da Silva, G. W. Critchlow and M. Figueiredo, J. Adhesion Sci. Technol. 22, 14771494 (2008). 22. SIMULIA ABAQUS Users Manual. Dassault Systmes Simulia Corp., Providence, RI, USA (2008). 23. E. Dragoni and A. Strozzi, ASME Trans. J. Tribology 110, 193200 (1988).

You might also like