Received 28 March 2007; received in revised form 25 June 2007; accepted 26 June 2007Fretting fatigueinvolves early crack growth up to a few hundreds of microns. Meanwhile, the propagation phase is the follow-on growth of cracks until structural failure. In the initiation phase, the cracks are often found to develop at an*Corresponding author.E-mail address: (B. Yang).Available online at Engineering Fracture Mechanics 75 (2008) 15071515/locate/engfracmech0013-7944/$ - see front matter C211 2007 Elsevier Ltd. All rights reserved.1. IntroductionSevere stress state emerges in the contact zone between two mating components, causing localized plasticdeformation and damage. If loading is cyclic, it becomes more hazardous, leading to crack initiation andgrowth 4,7,13,16,22,23. This damage process including crack development is referred to as fretting fatigue.It has been practically divided into two phases, namely, initiation and propagation phases, depending uponthe size of a crack that can be detected by available nondestructive evaluation techniques. The initiation phaseAvailable online 15 August 2007AbstractMotivated by experimental observations, we carry out a numerical analysis of the two-stage crack growth under frettingfatigue by using an ecient and accurate boundary element method. To start with, the variation of stress field during aloading cycle is analyzed. Various values of friction coecient in the contact zone are considered, which is shown to con-siderably aect the stress field. Then, by assuming crack initiation to occur in the shear mode, a surface-breaking crack isintroduced to the specimen at the location of highest shear-stress amplitude. The crack-tip stress intensity factors (SIFs)are calculated for various crack lengths and at various crack angles ranging from 25C176 to 45C176 about the contact surface. It isshown that, for a loading ratio of 0.5, the cyclic mode-II SIF amplitude decreases with increasing crack length, whilst itsmean value increases. It suggests that the (first-stage) shear crack would sooner or later become dormant, or switch toanother mode that can provide continuous support of growth. Then, the first-stage shear crack is manually kinked intoa second-stage opening crack, and the follow-on driving force is analyzed. It is shown that the kinking event is only favoredafter the first-stage crack has grown to a certain length. The present study thus provides insights in the mechanics of two-stage crack growth that has been frequently observed in a typical dovetail joint under fretting fatigue. It also suggests animproved experimental setup to quantitatively investigate the fretting fatigue in dovetail joints.C211 2007 Elsevier Ltd. All rights reserved.Keywords: Boundary element method; Contact mechanics; Crack initiation; Crack kinking; Dovetail joints; Fracture mechanics; Friction;Mechanics of two-stage crack growth in fretting fatigueB. Yanga,*, S. MallbaDepartment of Mechanical and Aerospace Engineering, Florida Institute of Technology, Melbourne, FL 32901, United StatesbDepartment of Aeronautics and Astronautics, Air Force Institute of Technology, Wright-Patterson AFB, Dayton, OH 45433, United Statesdoi:10.1016/j.engfracmech.2007.06.009inclination from the contact surface. Then, they kink and eventually propagate normal to the bulk tension,exhibiting a typical two-stage crack initiation and growth pattern 4,10,19,21.To characterize the fretting-fatigue crack initiation, Fellows et al. 5 utilized the shear-stress amplitude asthe key parameter in a stress-based approach. Lykins et al. 11,12 later showed through a combined numericaland experimental analysis with cylinder-on-flat and flat-on-flat contact configurations that the sites of fretting-fatigue crack initiation can indeed be reasonably explained with this parameter. Furthermore, Yang and Mall25 applied the crack analogue model/fracture-mechanics approach 6 to study the driving forces for initia-tion of mode-I and mode-II cracks at the edge of contact with friction coecient in the contact zone as the keyparameter. By comparing with experimental observations, they showed that the fretting-fatigue crack initia-tion was in the shear mode.On the other hand, the observed kinking of an initiated fretting-fatigue crack suggests a transition of frac-ture mechanism from a shear to an opening mode. A few fracture-mechanics-based approaches have been uti-initiation 5,9,18,22. These critical-plane-based approaches could be improved if a two-stage crack initiation1508 B. Yang, S. Mall / Engineering Fracture Mechanics 75 (2008) 15071515and early growth consideration is included. This has motivated the present study.In the present work, we numerically simulate the two-stage cracking process in fretting fatigue guided byexperimental observations. To start with, the variation of stress field during a typical loading cycle is analyzed.Various values of friction coecient in the contact zone are examined, which is shown to considerably aectthe stress field. Then, a surface-breaking shear-mode crack is introduced at the location of highest shear-stressamplitude in the contact zone. The crack-tip stress intensity factor (SIF) is calculated for various crack lengthsand various crack orientation angles ranging from 25C176 to 45C176 about the contact surface. It is then manuallykinked into an opening crack based on the criterion of maximum tangential-stress amplitude. This parametricstudy suggests that the crack should become dormant, or switch from the shear to an opening mode by kink-ing, after a certain distance of propagation. This is consistent with the experimental observations. In Section 2,the nonlinear problem of frictional contact and crack is formulated. It is solved by applying the boundary ele-ment (BE) method 1,24. In Section 3, the stress field in the contact zone before crack initiation is analyzed forvarious fretting-fatigue parameters including the loading ratio and the friction coecient. In Section 4, thetwo-stage process of fretting-fatigue crack initiation and early growth is analyzed. In Section 5, conclusionsare drawn.2. Problem formulationOne of the major concerns and considerations in the design of a dovetail joint, for example, used to connecta blade to a disk in gas turbine engines (Fig. 1a), is fretting fatigue. When a cyclic loading is applied, the forcelized to analyze fretting-fatigue crack initiation and growth, without separating them, under either opening-mode or mixed-mode condition along a prescribed path 2,14,15. However, a parametric study of the two-stage cracking process in fretting fatigue has not been reported in the literature. Meanwhile, a number of crit-ical plane-based approaches without any consideration of fracture-mechanics principles have been proposedto predict the crack-initiation life based on the stress/strain amplitude at the contact surface before the crackFig. 1. (a) A typical dovetail joint connecting a blade and a disk in a turbine engine. (b) A selected window area of (a) upon rotation.acting in the contact zone between the connected parts (blade and disk) varies in both normal and tangentialcomponents. The two components are coupled in a way determined by the geometry of the joint. However, itwas often experimentally studied with a fretting test setup, as well as analytically, where the normal force com-ponent (or normal displacement component) is held constant while the tangential force component is varied(3,4,9,16,17, among many others). The realistic geometry of a dovetail joint has also been taken into accountin several fretting-fatigue studies 8,18,20. In the present study, we consider a configuration where the normaland tangential loading components are coupled, as shown in Fig. 2. It represents more closely the realistic con-dition in a dovetail joint, as shown in Fig. 1b, a part of Fig. 1a. Furthermore, the setup may be realized in alaboratory as easily as the previous ones.The setup shown in Fig. 2 consists of two parts: a specimen, and a component to apply fretting loads to thespecimen through a pad. In our simulation, the specimen and loading component are taken to be the samematerial, and it is isotropic and linearly elastic. The specimen is restrained by a smooth sliding guide of a rigidwall along its bottom and right boundaries. The loading component is subjected to a normal traction r alongits left boundary, and is allowed to slide smoothly along the top boundary against a rigid wall. The loadingcomponent is initially in touch, through a cylindrical pad, with the flat top surface of specimen, but withoutpre-stress. Because of the wedge shape of the loading component, the pad and the specimen top surface comein contact over a finite zone when r is applied. The structure is free of traction otherwise. The boundary con-ditions and other details are shown schematically in Fig. 2. The loading direction, b, may be adjusted fordesired fitting tightness between the loading component and the specimen, which closely represent the partsand thelaw,B. Yang, S. Mall / Engineering Fracture Mechanics 75 (2008) 15071515 1509s fp_wj _wj; 1where s and p are the tangential and normal components of traction, f is the friction coecient, _w is the timerate of change of relative displacement between the surfaces in contact, and j _wj is its magnitude. The tractioncomponents, s and p are defined with the specimen.-1.2-0.8-0.400.40.8-1.2 -0.8 -0.4 0 0.4 0.8-0.0100.01-0.8 -0.4 0 0.4 0.8 (t)x/Ly/LTrailing edge of contactCrackSpecimenLoadingcomponentFig. 2. A setup simulating a dovetail joint (Fig. 1) under fretting fatigue. The inserts show a cylindrical pad of loading component and acrack possiblyspecimen. The tangential interaction between contact surfaces is modeled by a Coulomb-type frictionof a blade and a disk in a dovetail joint, respectively.When the setup (Fig. 2) is subjected to fatigue loading, i.e. cyclic r, a crack may nucleate at the contactsurface in the specimen. The crack may be opened, closed, or partially both during a loading cycle. Thenon-penetration condition of opposite surfaces is enforced upon a closed crack as well as between the padinitiating and growing at the trailing edge of contact. The dimensions are scaled by a characteristic length scale L.tration of elements in and around the contact zone. The following solutions had all been subjected to a con-xyindicatestial-stresinto tensicontactwell asTofor the1510 B. Yang, S. Mall / Engineering Fracture Mechanics 75 (2008) 15071515to the peak load and then unloaded to the half way down, i.e., loading ratio R = 0.5. Stresses along the contactsurface at the peak load and the final unloading point are plotted in Fig. 4a and b. Compared to the previouscase, a smaller contact zone with lower contact pressure ry, higher shear traction component sxy, and highertangentialon, showing a slight concentration, and finally diminishes. These figures show the characteristics ofstress variation during a loading cycle before crack initiation in the setup simulating a dovetail joint asthe accuracy of the numerical solution.understand the role of friction, another simulation was run with the same parameters as above exceptfriction coecient in the contact zone being changed to 0.7. The system was loaded up monotonicallyconcentration) at the leading edge of contact. Fig. 3bd shows the stress state after unloading. During theunloading process, the pressure, i.e. the normal traction component rydecreases. Reversed slip occurs at bothof the contact edges, with a stick zone inside. The stick zone recedes gradually and diminishes. The tangentialstress component rxat the trailing edge of contact (tensile at the peak load) quickly turns into compression.However, the tangential-stress component at the leading edge of contact (compressive at the peak load) turnsthat the contact zone is in the full slipping condition, i.e. under the gross slip condition. The tangen-s component rxis tensile and concentrated at the trailing edge of contact, but is compressive (withoutvergence check with mesh refinement. The final mesh is such that when the mesh density is doubled, therelative change of displacement is less than 0.1%.The first simulation was run with the system being loaded monotonically up to r = 0.005E, and thenunloaded completely. The friction coecient in the contact zone f = 0.3. Stresses along the contact surfaceof the specimen are recorded at the peak load and three intermediate unloading levels. The results are plottedin Fig. 3ad. In Fig. 3a, the system is subjected to the peak load. The variation of shear traction component sSince the material is linearly elastic (and isotropic), the displacement at a boundary point can be expressedas an integral of weighted displacement and traction along the boundary and crack. It is given for the loadingcomponent and for the specimen, respectively, as12uiZClduC3ijpjC0 pC3ijujC138dC; 212uiZCspuC3ijpjC0 pC3ijujC138dC ZCcrpC3ijwjdC; 3where C(ld)and C(sp)are the boundaries of the loading component and the specimen, respectively, C(cr)is oneside of a crack, u is the displacement, p is the traction, w is the displacement jump across a crack, and uC3ijand pC3ijare the fundamental solutions of displacement and traction of isotropic elasticity 1. To solve the crack prob-lem in a single-domain formulation, the following integral equation of traction is required,piZCspUC3ijpjC0 PC3ijujC138dC ZCcrPC3ijwjdC 4where UC3ijand PC3ijare combinations of derivatives of uC3ijand pC3ij, which can be found in 1 as well. The aboveEquations (2)(4) can be used to develop an ecient and accurate BE method to numerically solve the abovenonlinear boundary-value problem. One may refer to 1 for general details of the numerical technique, and24 for particular treatment of the nonlinear problem of frictional contact and crack. An iterative schemeis necessary to solve the present nonlinear problem.3. Contact stress analysisIn the following simulations, we set the loading direction b =30C176, radius of the cylindrical pad r =10L, andother geometrical parameters as shown in Fig. 2, where L is a length scale used to normalize all quantities ofthe dimension of length. A change in b or in r would aect significantly the magnitude but aect very little thecharacteristics of the stress field in the contact zone. The Youngs modulus E is used to normalize all quantitiesof the dimension of stress. The Poisons ratio m is set equal to 0.3. We adopted an adaptive mesh with concen-(tensile) stress rxconcentration at the trailing edge of contact are observed with higher value ofB. Yang, S. Mall / Engineering Fracture Mechanics 75 (2008) 15071515 1511-4E-30E+04E-3-4E-30E+04E-3xyxxyxij/friction coecient f in the contact zone. However, there appears to be little change in the qualitative featuresof the stress distribution along the contact surface.To prepare for the next analysis of two-stage crack growth, the amplitude of stress oscillation is examinedwith loading ratio R = 0.5 and friction coecient f = 0.3 and 0.7. Variations of the maximum normal-stressamplitude, the maximum shear-stress amplitude, and angle of the maximum shear-stress amplitude along thecontact surface are plotted in Figs. 5 and 6a and b, respectively. The angle of the maximum normal-stressamplitude is equal to the angle of the maximum shear-stress amplitude plus 45C176. First, it is seen that the trail-ing edge of contact experiences more severe fatigue loading than the leading edge of contact. Thus, the next-8E-3-0.12 -0.08 -0.04 0 0.04 0.08 0.12-8E-3-0.12 -0.08 -0.04 0 0.04 0.08 0.12-8E-3-4E-30E+04E-3-0.12 -0.08 -0.04 0 0.04 0.08 0.12-8E-3-4E-30E+04E-3-0.12 -0.08 -0.04 0 0.04 0.08 0.12yyxyxyxyxyx/Lx/Lij/Fig. 3. Variation of stress components (normalized by Youngs modulus E) along contact surface at (a) (peak load) r = 0.005 E;(b) (unloaded) r = 0.0035 E; (c) r = 0.002 E; and (d) r = 0.0005 E. In the case, f = 0.3.-1E-2-5E-30E+05E-31E-2-0.12 -0.08 -0.04 0 0.04 0.08 0.12-1E-2-5E-30E+05E-31E-2-0.12 -0.08 -0.04 0 0.04 0.08 0.12yxyxyxyxx/L x/Lij/Fig. 4. Variation of stress components (normalized by Youngs modulus E) along contact surface at (a) (peak load) r = 0.005 E and(b) (unloaded) r = 0.0025 E. In the case, f = 0.7.1512 B. Yang, S. Mall / Engineering Fracture Mechanics 75 (2008) 15071515-4E-30E+04E-38E-3-0.12 -0.08 -0.04 0 0.04 0.08 0.12x/Lf = 0.3f = 0.7/Fig. 5. Variation of maximum normal-stress amplitude along contact surface with friction coecient f = 0.3 and 0.7.4E-36E-32550f = 0.7 f = 0.7 f = 0.3deg)step of crack analysis will focus on the former. It is also seen that there are two peaks of the maximum normal-stress amplitude. There exists a higher peak at the edge of contact at the peak load, and a lower one within thereversed slip zone after unloading (at one half of the peak load). There is only one peak of the maximum shear-stress amplitude in the reversed slip zone. Most importantly, it is shown that both of the maximum normaland the maximum shear-stress amplitudes exhibit higher magnitude and higher gradient with greater valueof friction coecient in the contact zone. Thus, the resulting stress field would be increasingly hazardousfor crack initiation during fretting fatigue as the magnitude of friction coecient increases (due to surfaceroughening). In addition, the angle of the maximum shear-stress amplitude varies from about 5C176 to 45C176 withlocation from the reversed stick-slip boundary to the trailing edge of contact at the peak load.4. Crack analysisIn the section, we examine the process of two-stage fretting-fatigue crack growth in a dovetail-joint-likeconfiguration as shown in Fig. 2. The case of friction coecient in the contact zone equal to 0.7 and loadingratio equal to 0.5 that was studied before cracking in the previous section is c