Available online at www.sciencedirect.

com

International Journal of Fatigue 30 (2008) 15091528

International Journalof Fatigue

www.elsevier.com/locate/ijfatigue

A combined wear and crack nucleationpropagation methodology for fretting fatigue prediction
J.J. Madge, S.B. Leen *, P.H. Shipway
School of Mechanical, Materials and Manufacturing Engineering, University Technology Centre in Gas Turbine Transmission Systems, University of Nottingham, University Park, Nottingham NG7 2RD, UK Received 2 August 2007; received in revised form 5 November 2007; accepted 7 January 2008 Available online 20 January 2008

Abstract This paper presents a nite element based methodology for predicting the eects of fretting wear on crack nucleation and propagation under fretting fatigue conditions. The method combines wear modelling with critical plane, multiaxial prediction of crack nucleation and linear elastic fracture mechanics prediction of short and long crack propagation. A global-sub modelling nite element approach is employed for ecient wear and crack propagation simulation. The results are compared with available fretting fatigue test data and previous critical plane wear-based predictions (which neglect the propagation phase) for Ti6Al4V across a range of slip amplitudes. It is shown that the separate modelling of nucleation and propagation in the presence of wear can signicantly aect the prediction of fatigue life under both partial slip and gross sliding conditions. It is predicted that wear under gross sliding conditions signicantly retards propagation rate whereas wear under partial slip conditions is predicted to increase crack propagation rates across the slip zone. These results are consistent with observed phenomena of fretting fatigue cracking. 2008 Elsevier Ltd. All rights reserved.
Keywords: Fretting fatigue; Wear modelling; Crack propagation; Crack nucleation; Finite element; Life prediction

1. Introduction 1.1. General Fretting is a phenomenon which is associated with contacting surfaces nominally static with respect to each other, but which in fact, under operation, experience a relative sliding motion. This can lead to fretting wear in general and, in the presence of fatigue loads in one of the components, to fretting fatigue. Traditionally, fretting has been loosely described as corresponding to situations in which there is small relative motion, where small is often dened as being less than approximately 100 lm, but Varenberg has pointed out [1], via comparisons between macroscopic fretting tests and nano-scale fretting tests from an atomic force microscope, that since fretting can
*

Corresponding author. Fax: +44 0115 9513800. E-mail address: s.leen@nottingham.ac.uk (S.B. Leen).

occur in the latter with relative displacements of only about 5 nm to about 500 nm, a more specic denition is required. Varenberg has suggested that it is more precise to say that the relative motion surfaces should have a non-uniform distribution of local relative displacement at their contact [1]. In terms of avoiding confusion between reciprocating sliding (non-fretting) and gross sliding (fretting) cases, the latter has also proposed the concept of a slip index, and it was argued that when the slip index value is above 10, fretting conditions do not occur, i.e. reciprocating sliding pertains. Many engineering applications suffer from fretting, causing substantial reductions in operational life. Fig. 1 shows an example of fretting fatigue cracking in a laboratory scale spline coupling which occurred under an overload rotating bending moment loading [2]. The phenomenon is a complex one, and despite an increasing interest in the research topic, relatively little is known about the mechanisms that drive failure, particularly the interaction of the many variables which have

0142-1123/$ - see front matter 2008 Elsevier Ltd. All rights reserved. doi:10.1016/j.ijfatigue.2008.01.002

1510

J.J. Madge et al. / International Journal of Fatigue 30 (2008) 15091528

Fig. 1. Fretting fatigue cracking location for a laboratory scaled spline coupling under combined torque, axial and bending moment overload [2].

[14] can be predicted. Crack self arrest is an attractive feature in fretting fatigue where nucleation is commonly unavoidable under the severe stresses and surface damage at the contact interface. FDP methods cannot predict self-arrest, since they predict damage to continuously accumulate. However, a component that can be designed to create only self arresting cracks is inherently safe. Moobola et al. [14] state that to create self arrest in fretting, a relatively low fatigue load is required; indeed sometimes a bulk compressive load is required to suppress the action of contact stresses on crack growth. Of course, there will be some problems where highly stressed components are a necessity (e.g. due to space or mass limitations), so that it is not possible to achieve self arrest anyway. Previous investigations have tended to concentrate on either crack nucleation or crack propagation. One notable exception is the work of Araujo and Nowell [6], where an analytical approach was employed to solve for both the nucleation life (to a 1 mm crack) and the propagation life (to failure). 1.4. Wear eects Direct modelling of the nucleation phase is dicult, although one possible approach is the application of polycrystal plasticity techniques, e.g. [15]. However, this approach is computationally very intensive, particularly for application to industrial components and as a design tool, and the work of Goh at al. [15] was primarily focussed on ratchetting phenomena, which may be a contributory mechanism in crack nucleation. One objective of the present work is to develop a practical design tool for designing complex aeroengine components which experience fretting damage. Undoubtedly, wear damage plays a key role in crack nucleation under fretting fatigue conditions. Even if the mechanism of wear damage cannot yet be modelled directly, simulation of material removal due to wear can help to capture key phenomena related to both nucleation and propagation. One example is the observed dependence of fretting fatigue life on slip amplitude, particularly the recovery of life in the gross sliding cases. The classical mechanistic explanation of this phenomenon has been that the ablation of the surface serves to remove micro-cracks before they can nucleate and/or propagate. Recent nite element (FE) studies at the University of Nottingham [16,17] have combined a wear modelling technique similar to that developed by McColl et al. [18], which simulates the cycle-by-cycle evolution of the contact surfaces due to material removal, with a multiaxial SWT-based fatigue lifing approach, using a damage accumulation approach to account for the cycle-by-cycle evolution of stresses and damage. The method successfully predicted the measured eect of slip amplitude on fretting fatigue life for a cylinder on at Ti6Al4V contact, as reported by Jin and Mall [19], for a fretting test set-up as shown schematically in Fig. 2. It was shown that two phenomena are responsible

been observed to aect fretting fatigue life experimentally. Of the fty fretting variables that have been suggested, the primary variables are generally accepted to be: contact pressure, slip amplitude, coecient of friction and bulk stress level. However, even in terms of these primary variables, it is dicult to predict key phenomena. A number of dierent approaches have therefore been applied to develop understanding and predictive capability. Critical plane fatigue damage parameters and fracture mechanics techniques have proved to be relatively successful. 1.2. Critical plane parameter approaches Szolwinski and Farris [3] rst applied the critical plane approach to fretting. The method maximises a fatigue damage parameter (FDP) over a number of dierent planes and predicts life based on the most damaging plane. The SmithWatsonTopper (SWT) [4] and FatemiSocie (FS) [5] parameters have been used for life predictions in a number of studies with the critical plane approach [6,7], although others such as the modied shear stress range (MSSR) [8], and Crossland [9] parameters have also received attention. The Dang Van [10] parameter is a special case as it is argued to be microstructurally-based, but it does not provide a nite life prediction, only distinguishing between cases of failure and non-failure. 1.3. Fracture mechanics approaches A number of investigators have applied fracture mechanics to fretting fatigue problems [1113]. One benet of this approach is that stress gradient eects can be directly addressed, if crack growth is explicitly modelled as in [12], so that phenomena such as crack self arrest

J.J. Madge et al. / International Journal of Fatigue 30 (2008) 15091528

1511

Nf Ni Np

where Ni is the predicted number of cycles for nucleation of a short crack, typically about 10 lm depth, and Np is the predicted number of cycles for this short crack to grow to failure, which is dened here as the number of cycles at which the crack length becomes equal to half the fatigue specimen depth. The wear analysis is used to determine the evolution of both contact and subsurface stress and strain data, which are used to determine the nucleation life Ni and location. A fracture mechanics sub-model is employed to simulate crack growth from the predicted nucleation site. As described below, both short and long crack growth are included. Surface tractions are applied to the sub-model which mimic the contact stresses that are predicted by the wear model. Hence, the eect of wear on both nucleation and propagation can be investigated.
Fig. 2. Schematic of cylinder on at fretting fatigue test conguration.

2.2. Nucleation life prediction for the eect of slip on life. One is the removal of damaged material due to wear and the second is the redistribution of contact stresses, favourably in the case of gross sliding, unfavourably in the case of partial slip, due to the geometry evolution. Fig. 3 shows the contrasting cases of (a) gross sliding where the peak contact pressure is gradually reduced and the contact patch increased with increasing numbers of fretting cycles and (b) partial slip where a localised peak develops at the stick-slip interface. However, this work did not distinguish between crack nucleation and propagation. Therefore, the purpose of the present work is to present an FE-based fretting fatigue prediction methodology which combines the eects of cyclic wear simulation with crack nucleation and propagation prediction, including short crack growth prediction. The advantage of the FE basis is that it facilitates application to more complex geometries for more direct application to industrial design problems, e.g. [20]. The new methodology will permit prediction of the following key phenomena, which cannot be captured using each technique in isolation: (1) (2) (3) (4) Crack self arrest. Dependence of fretting fatigue life on slip amplitude. Eect of wear on crack growth. Relative importance of nucleation and propagation behaviour. Nucleation is modelled using a critical-plane implementation of the SWT fatigue damage model, described in detail in [17]. According to the SWT model, for any given loading cycle, dened by a maximum strain range over one cycle, De, and a peak normal stress rmax, on any given plane orientation, the predicted damage per cycle is 1/Ni, where Ni is dened by SWT rmax De r02 2b bc f 2N i r0f e0f 2N i 2 2 E where r0f is the fatigue strength coecient, E is Youngs modulus, b is the fatigue strength exponent, e0f is the fatigue ductility coecient and c is the fatigue ductility exponent. An inconsistency in the use of Eq. (2) for predicting nucleation of a 10 lm crack relates to the fact that the fatigue constants on the right hand side are normally obtained from fatigue tests where the associated number of cycles is taken as either corresponding to specimen failure or detection of a 1 mm crack. This is discussed further below, where a method is presented to redress this inconsistency. Due to the evolving nature of the stressstrain history from cycle to cycle, as a consequence of material removal (see Section 2.4) and the associated stress redistributions, it is necessary to employ a damage accumulation framework. As in [17] Miners rule is employed, whereby nucleation is dened to have occurred at a material point when the total accumulated damage x reaches a value of 1, where x is dened as x
Ni X 1 N i;j j1

2. Methodology 2.1. Overview Fig. 4 shows a schematic representation of the proposed new methodology, which combines wear modelling with crack nucleation and crack propagation prediction. The total fatigue life is dened as

where Ni,j is the critical-plane SWT predicted number of cycles to failure for fretting cycle j (found using Eq. (2)) and Ni is the total number of fretting cycles to nucleation, which is the key output from this calculation. This method can also provide information on the orientation of the nucleated crack.

1512

J.J. Madge et al. / International Journal of Fatigue 30 (2008) 15091528

Fig. 3. Contact pressure evolutions due to wear for: (a) gross sliding for calculated contact slip (d) of 8.7 lm, and (b) partial slip for d of 2.8 lm [17].

2.3. Crack growth The stress intensity factor K is used to characterise the crack tip stress eld. As mentioned, the contour integral method is used to assess the stress intensity factors for both the opening mode (K1) and the sliding mode (KII) around each fretting cycle, furnishing a maximum and minimum value for any given cycle analysed. The eective stress intensity factor range DKe is then found [21] DK I K I;max K I;min where K I P 0 4 5 6 DK II K II;max K II;min q DK eff DK 2 DK 2 I II

eral application. The version used here makes use of modications which allow better description of near thresholdand short crack growth behaviour. Cracks are found to self-arrest if the stress intensity factor range is below some critical threshold DKth. Short cracks are often found to propagate at rates signicantly higher than a long crack with an equivalent SIF. In addition short cracks can propagate at stress intensity factor ranges below DKth. El-Haddad et al. [22] proposed that the growth of a short crack of length l can be described by analysing the crack as if it were actually of length (l + l0) so that p 7 DK QDr1 pl l0 where Q is the geometric form factor, Dr1 is the far eld stress and l0 is a ctitious crack length which is found to correspond with the Kitagawa and Takahashi [23] critical crack size from threshold fatigue behaviour and is dened as follows:

The well-known Paris law can be used to correlate crack growth rate with stress intensity factor range. A range of developments have been suggested for the Paris crack growth law since its original conception to allow more gen-

J.J. Madge et al. / International Journal of Fatigue 30 (2008) 15091528

1513

Fig. 4. Schematic showing three stages in the proposed wearnucleationpropagation fretting fatigue methodology.

l0

 2 1 DK th p re

where 1 3 4m cos h1 cos h 16pG 1 c12 sin h cos h 1 2m 8pG 1 c22 41 mcos h 1 2m1 cos h 16pG 3 cos h 11 cos h c11 12 13

where re is the fatigue limit. As a nite element formulation is used to determine the stress intensity factor, it is necessary to approximate the ElHaddad stress intensity factor DKe,EH as a perturbation of the FE determined value DKe, as follows: DK eff;EH r l l0 DK eff l 9

14

This approach means that the eective threshold SIF below lo increases linearly with increasing (short) crack length. The crack growth law used in this study is thus formulated ! r!m dl l l0 m C DK eff K th 10 dN l
dl where dN is the cyclic growth rate, C and m are the Paris constants. In a study specically aimed at comparatively assessing a number of dierent approaches for modelling short and long crack growth in fretting fatigue, Navarro et al. [24] concluded that Eq. (10) gave the best correlation with measured life, compared with a number of alternative crack growth laws for fretting fatigue of Al-7075. Under the multiaxial stress state created by fretting loads it is possible for cracks to change direction. A number of dierent criteria have been suggested in the literature. The approach used here is the Sih [25], minimum strain energy density function. For two-dimensional problems the strain energy density function S is given by

with h as the angle between the existing crack propagation direction and the new propagation direction, m is Poissons ratio, taken here as 0.32, and G is the shear modulus. h is selected such that S forms a local minimum. If more than one minimum occurs within the domain p < h < p, the minima chosen is that which gives the maximum absolute value of S [26]. 2.4. Wear modelling The geometry evolution of the contact due to wear is modelled using a modied version of the Archard equation [17] Dhx; t k px; tdx; t 15 where Dh(x, t), p(x, t) and d(x, t) are the incremental wear depth, contact pressure, and relative slip at a node at horizontal position x, where the x-axis origin is at the centre of the contact, and time t, respectively, and k is the wear coefcient. To model each individual cycle would be too computationally expensive, so a cycle jumping technique is employed. Wear is assumed to be linear over a given cycle

S c11 K 2 c12 K I K II c22 K 2 I II

11

1514

J.J. Madge et al. / International Journal of Fatigue 30 (2008) 15091528

jump DN. In this way one FE cycle simulates the cumulative wear eect of DN cycles Dhx; s DNkpx; sdx; s 16 where s is the time within one cycle corresponding to DN wear cycles. The spatial adjustment of the contact nodes is achieved via a user subroutine called UMESHMOTION and within an adaptive meshing framework in the general purpose, non-linear FE code ABAQUS [27]. When this cycle jumping technique is combined with Eq. (3), the nucleation damage rule becomes
DN X DN x N i;k j1 Nt

17

where Nt is the total number of fretting cycles simulated and Ni,k is the critical-plane SWT predicted number of cycles to failure for fretting cycle-jump k. Given the adoption of a critical-plane approach to calculate fatigue damage, it is necessary to make an assumption about how damage on a given plane interacts with damage on other planes from cycle to cycle. One approach is to calculate the critical plane SWT for each wear increment and simply accumulate this value from wear increment to increment, irrespective of the fact that the associated orientations of the critical plane may change with wear, i.e. assume the critical-plane SWT damage to have an isotropic damaging eect. An alternative approach is to calculate SWT values for all plane orientations for each wear increment and to accumulate the incremental damage on each plane. The latter approach is signicantly more computationally intensive, requiring

storage of damage on 36 dierent planes at each element centroid (note that a rectangular grid of 1700 elements is monitored for damage here) for each wear increment (typically about 100 wear increments). A more rigorous alternative again would be an anisotropic damage mechanics approach, which could include interaction eects between damage on dierent planes, but this is beyond the scope of the present paper. Due to the use of adaptive meshing to update the mesh on removal of material, specic elements and nodes are no longer linked uniquely to actual material points throughout the analysis. Consequently, a material point mesh (MPM) is created as the global reference for damage accumulation; the nodes of the MPM have xed coordinates throughout the analysis. Cyclic damage is calculated at the centroid of each element, over an area corresponding to 1.1a0 into the contact depth and 4a0 across the contact width (a0 is the initial contact semi-width for an unworn geometry) and linearly interpolated back to the MPM for accumulation. In this way, nodes on the MPM corresponding to removed material (due to wear) do not accumulate any further fatigue nucleation damage. 3. Finite element modelling 3.1. Overview A global-sub model approach is employed to solve the crack propagation problem in the presence of wearinduced geometry modication. Fig. 5 shows the global

Fig. 5. Schematic showing transfer of traction data from global wear model to crack submodel.

J.J. Madge et al. / International Journal of Fatigue 30 (2008) 15091528

1515

model and crack growth sub-model of the fretting fatigue test arrangement, as discussed in more detail by Madge et al. [17], and based on the experimental work of Jin and Mall [8], the specimen geometry of which is shown in Fig. 6. The time histories of contact pressure p(x, t) and shear traction q(x, t) are calculated using the global model over Nt fretting cycles, along with crack nucleation data for input to the SWT critical plane calculations. p(x, t) and q(x, t) are then passed to the sub-model and crack growth is simulated over Nt fretting cycles. Fig. 7 shows a more detailed owchart of the methodology, which has been automated via a MATLAB program, created to drive the crack growth analysis, in conjunction with ABAQUS. 3.2. Material, loading history and model details As shown in Figs. 2 and 6 [8], a pair of cylindrical (radius 50.8 mm) fretting pads are held in contact with a at, uniaxially-loaded fatigue specimen (depth 3.8 mm). The material used in the experiments is a dual phase Ti6Al4V alloy which consists of a hexagonal close packed and b body centred cubic phases. The material was solution heat treated at 935 C for 1.75 h, cooled in air and annealed at 700 C for 2 h in vacuum and cooled in argon. A symmetry plane parallel to the specimen axis is employed in the global FE model so that only one pad needs to be modelled (Fig. 5). A material model of Ti 6Al4V with a Youngs modulus of 126 MPa, a yield stress of 930 MPa and a Poissons ratio of 0.32 was used. Fig. 8 shows the load cycle history for the rst few load cycles. In the rst analysis step a normal load P of 208 N/mm is applied to the cylinder, resulting in a peak Hertzian contact pressure of 302 MPa. In the next step the specimen is loaded by a cyclic fatigue load r(t) with a maximum value rb of 550 MPa and a stress ratio R of 0.03. The pad is also loaded with a prescribed maximum displacement dapp,m, which is varied for each simulation to cover a range of slip amplitudes including partial slip through to gross

sliding conditions. Note that dapp(t) is the distance by which the fretting pad is displaced relative to the global coordinate axis. The slip range d is the maximum slip observed on the contact surface over one cycle. To model the eect of wear, dapp(t) and r(t) are applied cyclically in the FE model. Using the von Mises yield criterion, the macroscopic material behaviour is predicted to remain elastic throughout the loading history. The Lagrange multiplier contact algorithm was used to strictly enforce the stick condition when the shear stress is less than the critical value according to the Coulomb friction law. In the fretting experiments, the coecient of friction (COF) typically starts low, and rises to a higher value within the rst few thousand cycles (Jin and Mall [8]). Sabelkin and Mall [28] have suggested that a constant COF value of 0.8 is representative for the tests studied here. A value for wear coecient k of 2.75 108 MPa1 has been estimated as representative of the Ti6Al4V fretting contact pair of the present work (see Madge et al. [17] for details), using measured wear scar data from Magaziner et al. [29]. The values of the Paris constants C and m employed here are: C = 1.25 10 11 MPa2.59 mm0.295 p and m = 2.59 [6]. DKth is taken as 133 MPa mm [12]. A range of dierent l0 values between 10 and 60 lm have been suggested for Ti6Al4V at dierent stress ratios [30]. In this study the value of l0 is taken as 20 lm, based on values of 133 MPa mm1/2 and 569 MPa for DKth and re, from [31], following the approach used in [32]. The accuracy of the global FE model has been validated in [17] against available theoretical solutions for both partial slip and gross sliding. The FE-implementation of the critical-plane SWT parameter follows the method described by Sum et al. [7]. In practice this is achieved by transforming the time histories of element centroidal stresses and strain ranges onto planes at 5 intervals over a 180 range using the twodimensional transformation (Mohrs circle) equations for stress and strain. The maximum normal stress rmax with respect to time, and the corresponding strain range De are determined for each of the 36 planes in each element. De is the dierence between the maximum and minimum values of strain normal to the candidate plane over the complete loading cycle. Thus, SWT values are obtained for each candidate plane in each element. These values are then employed to establish the maximum critical plane SWT value with respect to plane orientation in each element, which in turn is used with Eq. (2) to furnish a number of cycles to nucleation, Ni. The SWT implementation has been successfully validated against corresponding FEbased predictions by Sum et al. [7] and volume-averaged results of Araujo and Nowell [6] for partial slip fretting conditions of an Al4%Cu alloy. 3.3. Combining critical plane and LEFM approaches

Fig. 6. Schematic of: (a) the fretting fatigue specimen, and (b) fretting pad. After Jin and Mall [8].

Table 1 shows the Ti6Al4V SWT constants available from Dowling [33], which have been derived from fatigue

1516

J.J. Madge et al. / International Journal of Fatigue 30 (2008) 15091528

Fig. 7. Flow chart showing computational sequence for submodel analysis.

P
P
b

Table 1 SWT constants for Ti6Al4V corresponding to a 1 mm crack (from Dowling [33]) r0f (MPa) b 0.104 e0f 0.841 c 0.688

(t) app(t)
Time

2030

app,m

min

Fig. 8. Normal load, tangential displacement and substrate fatigue stress histories implemented.

tests where failure is dened as the occurrence of a 1 mm crack. However, the length scales of damage relevant to crack nucleation in the present work are signicantly

shorter than this; specically, a crack is assumed to have nucleated here (i.e. x = 1) when it has reached a depth of 10 lm. The element depth at the surface is 16 lm, so that the centroid, where damage is monitored is at 8 lm. This assumed to be suciently close to 10 lm to assume a 10 lm crack, with negligible eect on the resulting predictions. Hence a modied set of fatigue constants is derived corresponding to 10 lm nucleation. The SWT equation is

J.J. Madge et al. / International Journal of Fatigue 30 (2008) 15091528

1517

based on the combination of the Basquin high cycle fatigue (HCF) equation, for stress to life relation, and the Con Manson low cycle fatigue (LCF) equation, for plastic strain to life relation, as follows, respectively Dr b r0f 2N f 2 Dep c e0f 2N f 2 18 19

with consideration of the peak stress to account for the mean stress eect. The approach adopted here is as follows:

1. Establish Nf across the 300500 MPa range of stress amplitudes, corresponding to a 1 mm crack length, using the Basquin constants of Table 1. 2. Estimate Np, the numbers of cycles for a 10 lm crack to grow to 1 mm, for a series of stress levels across the range 300500 MPa, using LEFM with the El-Haddad correction for short crack growth, using the solution for a semi-elliptical surface crack in a round bar under tension from [34] p K I F Ir l 20 where FI is taken to correspond to the position of maximum SIF along the crack-front. 3. Calculate Ni = Nf Np across the same range of stress amplitude levels and hence obtain the new Basquin constants corresponding to a 10 lm crack. Table 2 shows the resulting modied Basquin constants and Fig. 9 shows the 10 lm nucleation and the 1 mm crack stress-life curves.

Table 2 Estimated Basquin constants corresponding to 10 lm crack nucleation in Ti6Al4 V r0f (MPa) 1817.2 b 0.0978

550 500 450

3.4. Stress intensity factor determination The model primarily uses modied pressure quadratic triangular elements, as these provide a more versatile remeshing capability than quadrilaterals for potentially irregular crack paths. The meshes are generated using the DistMesh Matlab suite [35]. The contour integral facility within ABAQUS is used to evaluate the stress intensity factors; to this end, a rosette of quadratic elements is implanted at the crack tip to enable crack tip parameter evaluation (Fig. 10). For the rosette elements a quadratic quadrilateral formulation is used, the elements sharing the crack tip node use a p quarter-point node spacing to provide the correct 1/ r strain singularity for LEFM, where r is radial distance from the crack tip. Figs. 11 and 12 show that this approach gives very good agreement with benchmark solutions from

S (MPa)

400 350 300 250 200

1 mm crack 0.01 mm crack

log 2N i

Fig. 9. Comparison of Basquin stress-life curves corresponding to nucleation of a 10 lm crack and development of 1 mm crack.

Fig. 10. An example mesh showing initial 10 lm surface defect and detail of rosette of quadrilateral elements around the crack tip.

1518

J.J. Madge et al. / International Journal of Fatigue 30 (2008) 15091528

Fig. 11. Comparison of theoretical [36] and FE-based predictions of crack propagation in SENT specimen for transverse crack (plain fatigue).

the literature, for a range of slanted cracks in SENT specimens under plain fatigue [36,37]. 3.5. Contact stress evolution treatment The impact of wear on fretting fatigue crack growth is studied here through the eect of evolving stresses on propagation. In the cases studied here the crack propagation rate is signicantly higher than the wear rate, so that the reduction in length of a crack due to wear is negligible. The stress evolution aspect of wear is investigated by updating the sub-model boundary tractions p(x, t) and q(x, t) derived from the global model wear analysis. From a computational viewpoint, it is desirable to use large cycle-jump sizes for the wear simulation, in this case DN of 1500 cycles, whilst ensuring that accuracy and stability are not compromised. However, the forward prediction technique used in the crack propagation simulation requires a much smaller cycle-jump size, e.g. 100 cycles. Consequently, a linear interpolation scheme was implemented to determine the pressure and shear traction distributions for each crack growth cycle-jump, which are intermediate to the wear cycle-jump p(x, t) and q(x, t) data available from the global wear model. 4. Results 4.1. Comparison with test data Fig. 13 shows the eect of slip dapp,m on fretting fatigue life, as obtained from: (i) the present methodology (referred to as wearnucleationpropagation), which separately

predicts nucleation of a 10 lm crack and propagation of that crack to specimen half-depth, including the eects of geometry evolution due to wear, (ii) previous life predictions from [17] (referred to as wear-SWT total life), which did not distinguish between nucleation and propagation, but did include the eects of geometry removal due to wear and predicted life to a 1 mm crack, (iii) the predicted life to nucleation of a 10 lm crack (referred to as wearnucleation) including the eects of geometry evolutions due to wear and (iv) the measured lives from the work of Jin and Mall [8]. Note that the results of [8] have been corrected in order to relate the remote slips recorded experimentally to the more local slip dapp,m, as described in [17] and based on the relationship presented by Sabelkin and Mall in [28]. Further, due to two slightly dierent apparatus being used in [8], only those results obtained using a constant normal load are considered here as this matches the boundary condition used in the model more closely. Fig. 14 shows an example of the wear depth evolution for the fatigue specimen under gross sliding conditions with an applied displacement dapp,m of 8.7 lm. 4.2. Eect of nucleation period The combined nucleationpropagation predictions of Fig. 13 are predicated on the assumption of an initially perfect material, with no inherent aws, so that the nucleation life is dened as the number of fatigue cycles from zero (initial) damage to a damage level equivalent to a 10 lm crack. However, real materials will not be perfect but will have initial aws and weaknesses which can be modelled as initial damage. One of the benets of the present approach is

J.J. Madge et al. / International Journal of Fatigue 30 (2008) 15091528

1519

Fig. 12. Comparison of benchmark solution [37] and FE-based solutions for SIF for slanted (angle shown is angle of crack to specimen surface) cracks in SENT specimens for: (a) Mode I, and (b) Mode II. (Note: curves show benchmark solution, crosses show authors FE data).

the ability to separate the nucleation and propagation lives, so that the nucleation life can be varied to simulate dierent numbers of cycles to nucleation. As the contact stresses continuously evolve due to material removal eects, the crack propagation life will be aected by the assumed (or calculated) nucleation life. Therefore the eect on subsequent fretting fatigue crack propagation of varying Ni is investigated here; this can be thought of as considering different degrees of inherent damage, so that for Ni = 0, for example, the material is assumed to already have a 10 lm crack present, whereas for Ni = 10,000, the assumption is that the 10 lm crack has occurred after 10,000 cycles, viz. the nucleation response of the material has been varied accordingly, or alternatively, inherent damage in the material is assumed so that the additional number of cycles required for a 10 lm crack is Ni = 10,000. Thus, for

Ni = 0, the crack is assumed to have nucleated immediately and initially grows under the inuence of the unworn surface tractions, whereas for Ni = 10,000, for example, the initial aw grows in a worn contact stress eld relating to the 10,000th cycle. In the present work, two dapp,m values have been studied, one corresponding to an applied displacement dapp,m of 8.7 lm and another corresponding to dapp,m = 2.9 lm. The FE-predicted contact slip distributions corresponding to both of these cases have been given in [17], where it is clear that the dapp,m = 2.9 lm case gives rise to a central stick region from about x = 0.027 mm to about x = 0.1 mm, with non-zero slip elsewhere reaching a maximum value of about 2.9 lm at one contact edge (x is distance from the centre of contact). In contrast, the dapp,m = 8.7 lm gave no stick region, but still gave a

1520

J.J. Madge et al. / International Journal of Fatigue 30 (2008) 15091528

Fig. 13. Comparison of predicted fretting fatigue life as a function of local slip amplitude, dapp,m, (see Fig. 5) against test data of [8], including results of wearnucleationpropagation method, along with wearnucleation and wear-SWT total life method of [17].

Fig. 14. Predicted evolution of wear depth distribution on fatigue specimen for dierent numbers of fretting cycles for the dapp,m = 8.7 lm case.

signicantly non-uniform distribution on contact slip, again reaching a maximum value of about 8.7 lm at one contact edge. It should be noted that in the present work (see Fig. 5) the point of application of the applied displacement is very close to the contact surface. Varenberg [1] has pointed out the importance of understanding the relative dierences between applied displacement and contact slip, in assessing which fretting regime pertains in a given situation. To this end the concept of a slip index was introduced,

with a view to uniquely dening the fretting regime and the partial slip and gross sliding fretting regimes, with identication of each regime (and hence the associated values of slip index) being based on the observed friction loops. As discussed in more detail in [17], the two cases of gross sliding and partial slip studied here, correspond directly to two cases in [8], which have been identied as gross sliding and partial slip also via the associated characteristic friction loops.

J.J. Madge et al. / International Journal of Fatigue 30 (2008) 15091528

1521

Fig. 15. Predicted eect of nucleation time on the gross sliding propagation behaviour (xi/a0 = 1.02).

Fig. 15 shows the predicted eect of varying Ni on the subsequent crack propagation under gross sliding conditions. If nucleation is assumed to occur instantly the propagation period is 4300 cycles, whereas if nucleation is judged to have occurred at 1 105 cycles, the total propagation time is 7600 cycles. This is longer than that found in the plain fatigue case of the same bulk stress load. Fig. 16 shows that towards the nal stage of propagation, the presence of the contact slows growth appreciably compared to the plain fatigue case. This is attributed to the widening of the contact width with wear, such that the crack is eventually under the contact and experiencing compressive stresses over an increasing portion of the loading cycle. It is important to note that the propagation period is small in comparison to the total life ($1%) for these gross sliding

cases, so that for these cases, the nucleation period is the more important consideration in terms of fretting fatigue life prediction. Fig. 17 shows the predicted eect of varying Ni for the partial slip case, for two dierent assumed nucleation locations, namely at the stick-slip interface and at the edge of contact, while Fig. 18 shows the eect of varying Ni across the complete (evolving) slip zone for the partial slip case. If the nucleation period is assumed to be negligible, i.e. Ni = 0, the propagation life is observed to be a relatively strong function of the nucleation location, with the edge of contact found to provide the most rapid propagation. However, as wear is allowed to advance by increasing the assumed nucleation period, the propagation life is found to become a much weaker function of the nucleation loca-

Fig. 16. Comparison of predicted gross sliding crack propagation rates with dierent nucleation periods (xi/a0 = 1.02) against plain fatigue growth prediction.

1522

J.J. Madge et al. / International Journal of Fatigue 30 (2008) 15091528

Fig. 17. The predicted eect of varying the nucleation period upon partial slip propagation behaviour at dierent cracking locations: (a) xi/a0 = 0.57, (b) xi/a0 = 1.02.

tion, i.e. propagation in the slip region has been accelerated signicantly by wear, specically the development of the stick-slip interface pressure peak (see Fig. 3b). Conversely, the eect of wear at the edge of contact is to retard propagation slightly, which is also consistent with the reducing pressure at the initial contact edge (Fig. 3b). Under partial slip conditions wear creates a wider region over which cracking could be expected, which helps to explain why cracking is often observed at multiple points along the slip region in experimental studies. 4.3. Eect of bulk stress: crack arrest In the experimental validation, the cases reported by Jin and Mall [8] use a relatively high bulk fatigue load of 550 MPa. The results presented here suggest that under

these high stress conditions the majority of the fatigue life is spent in the initiation phase. The phenomenon of crack arrest arises when the stress intensity at the crack tip drops below the threshold value for growth (Eq. (10)). This can occur either due to the crack having grown into a lower stress region or due to the stresses on the crack tip reducing due to stress redistribution. The sub-modelling approach adopted for the crack propagation analysis obtains the evolving contact tractions from the global model, which explicitly models the geometry removal due to wear. It is therefore possible to decouple the evolving contact tractions p(x,t) and q(x, t) from the bulk fatigue stress r(t), to explore the eect of varying the bulk stress under a constant surface traction evolution. In order to investigate this eect and thereby illustrate some key fretting fatigue phenomena, a wear-induced con-

J.J. Madge et al. / International Journal of Fatigue 30 (2008) 15091528

1523

Fig. 18. Partial slip predicted crack propagation life as a function of crack nucleation position.

tact stress evolution associated with a 550 MPa peak bulk fatigue load, is employed with two slip amplitudes; dapp,m = 8.7 lm (gross sliding) and dapp,m = 4.2 lm (partial slip). Each case is studied with a varying bulk fatigue stress applied to the sub-model crack propagation analysis. The propagation behaviour is then investigated using with wear and without wear analyses. The without wear analyses use the unworn pressure distribution, whereas the with wear analyses use the evolving surface tractions, as predicted by the global model. For each case the crack is placed at the point of maximum damage as predicted by the nucleation model. For the gross sliding case this is at the edge of contact and for the partial slip case it is at the edge of contact for the without wear analyses and at the stick-slip interface (xi/a0 = 0.121) for the with wear analyses. Table 3 summarises the results of the gross sliding cases. Generally, wear has the predicted eect of slowing crack growth compared to when wear is neglected; this is demonstrated at the 200 MPa fatigue load. However, if the fatigue load (rb) is dropped to 100 MPa, the role of wear becomes critical in failure prediction; Fig. 19 compares the crack growth curves for the with wear and without wear assumptions. The without wear assumption predicts that the crack propagates to failure, whereas the with wear assumption causes sucient attenuation of stresses that the crack arrests at a length of 26 lm after 1.3 104 cycles, having grown out of the most severe high stress region local to the contact surface. If rb is reduced to zero it is found that both assumptions lead to predictions of no crack growth. Table 4 shows the results of the partial slip cases. In the 0 MPa case, neither the with wear nor without wear analyses predict failure. However, Fig. 20 shows that the with wear analysis predicts that the crack is ini-

Table 3 Eect of bulk stress variation on gross sliding case results Bulk stress (MPa) 0 100 200 With wear No wear With wear No wear With wear No wear Final length (mm) 0.01 0.01 0.026 1.8 (Failed) 1.8 (Failed) 1.8 (Failed) Comments No growth No growth Arrested 13,000 cycles Failed 115,000 cycles Failed 40,000 cycles Failed 35,000 cycles

tially dormant, until about 4 104 cycles, at which point the contact stresses have evolved to a condition, which is favourable for crack growth, after which the crack grows to a length of about 0.8 mm, at which point it arrests, due to it having grown out of the fretting fatigue contact stress eld. When the bulk stress is increased to 50 MPa, the no wear analysis predicts no growth, i.e. innite life. However, as shown in Fig. 21, the with wear analysis predicts that although crack growth is relatively slow, the pressure peak which develops at the stick-slip interface is enough to drive the crack through to failure. At the higher stress level of 100 MPa, both analyses predict failure, after 1.51 105 cycles for the without wear model, and after 1.06 105 cycles for the with wear model. 5. Discussion In the experimental cases modelled, the predicted propagation lives are short when compared to the total expected life, and they are similar for both gross sliding and partial slip, even though the stress gradients are signicantly steeper and the peak stresses are higher for the worn partial slip cases. In the gross sliding cases the

1524

J.J. Madge et al. / International Journal of Fatigue 30 (2008) 15091528

Fig. 19. Comparing with wear and without wear predictions of crack length vs. number of cycles for gross sliding case with dapp,m = 8.7 lm and rb = 100 MPa.

Table 4 Eect of bulk stress variation on partial slip case results Bulk stress level (MPa) 0 50 100 Analysis type With wear No wear With wear No wear With wear No wear Final length (mm) 0.8 0.01 1.8 (Failed) 0.01 1.8 (Failed) 1.8 (Failed) Comments Reactivated 30,000 cycles, arrested 400,000 cycles No growth Reactivated 30,000 cycles, failed 364 000 cycles No growth Failed 151 000 cycles Failed 106 000 cycles

Fig. 20. With wear predicted crack growth curve for dapp,m = 4.2 lm partial slip case with rb = 0 MPa.

J.J. Madge et al. / International Journal of Fatigue 30 (2008) 15091528

1525

Fig. 21. With wear predicted crack growth curve for dapp,m = 4.2 lm partial slip case with rb = 50 MPa.

propagation life is typically only around 12% of the fatigue life. The rapid propagation of fatigue cracks under this loading regime is consistent with the plain fatigue results: simple 2D theoretical predictions of crack propagation life show that reducing the fatigue load from 550 to 250 MPa would increase life from 7500 to 80,000 cycles for a similar single-edge notched tension (SENT) geometry. In contrast, for the partial slip cases and the low life, low slip gross sliding cases, the predicted propagation life forms a greater proportion of the total life, e.g. up to about 20% for a slip of about 3.5 lm. Recent work on crack growth prediction in Ti6Al4V for fretting fatigue by Fadag et al. [31], without considering the eects of wear, has analysed the proportion of crack propagation as a function of load, assumed initial crack length and for both at and cylindrical pad fretting pad geometries. It was shown that, for the intermediate and high cycle regimes (with total life greater than about 105 cycles), predicted propagation life formed only a small part of total life (less than 10%). This was reported to be in agreement with experimental observations from [38] where the initiation and propagation proportions were estimated from post-mortem microscopic observations of striations. Clearly, these latter results are consistent with the predicted propagation proportions of Fig. 13. Fig. 13 shows that the wearnucleationpropagation methodology predicts the same general trend with respect to slip amplitude as the results of Madge et al. [17] (which are wear-SWT total life results) and the experimental data; specically, a critical range of slip amplitudes is predicted corresponding to the partial slip regime, which results in short fatigue lives. However, some dierences

are apparent between the results of the present methodology and those of the wear-SWT total life approach. In the partial slip region, the wearnucleationpropagation method is more accurate than the wear-SWT total life method, whereas in the gross sliding region, the converse is true. Figs. 15 and 17 indicate that the propagation period is less sensitive to slip amplitude than the initiation period. Fig. 13 (wearnucleation curve) shows that initiation life can vary by an order of magnitude with slip amplitude, whereas the maximum variation in propagation life is only around a factor of 2.0. In the dapp,m > 1 lm partial slip regime the reduction in initiation life through the adoption of short-scale constants is more than oset by the associated propagation life, so that the wearnucleationpropagation methodology predicts longer life than the wearSWT total life approach. In the gross sliding regime, the propagation life is small with respect to the reduction in nucleation life caused by the use of short-scale fatigue parameters. Hence, under gross sliding conditions, the wearnucleationpropagation methodology results in a smaller predicted life than the wear-SWT total life approach of Madge et al. [17]. Due to the complexity of the interactions between wear, nucleation and propagation, it is dicult to identify precisely why this is the case. One plausible explanation is that the 10 lm nucleation life is under-predicted, e.g. due to the simplifying assumptions made in Section 3.3, and this, compounded with the sensitivity of predicted nucleation life to slip amplitude, results in the dierences observed in Fig. 13. It is also worth pointing out a number of other related issues, as follows:

1526

J.J. Madge et al. / International Journal of Fatigue 30 (2008) 15091528

 The ConManson terms have not been modied in the process of identifying the short-scale SWT constants. The 10 lm nucleation life may thus be underestimated, since the high cycle fatigue crack propagation behaviour is thus assumed to apply across the full range of stress levels.  Crack closure and mean stress eects are not included in the crack propagation model; this will also contribute to dierences between the wear-SWT total life and the wearnucleationpropagation life.  The sub-model (crack propagation) analyses of this paper, do not explicitly model the change of geometry associated with wear, but only the associated (global model) predicted surface traction evolution. Consequently, the predicted (sub-model) surface stress evolution with fretting cycles will be slightly in error.  The use of a symmetry assumption means that it is assumed that there are two cracks growing at the same time from both sides of the specimen. This is not unreasonable when the cracks are short, since multiple crack initiation sites are typical of fretting experiments, but as the cracks become longer, e.g. within linear elastic fracture mechanics regime, this assumption may become less realistic, since typically only one dominant crack grows to failure. However, comparison of SENT and double-edge notched tension (DENT) SIF expressions suggests that for short cracks, l/W < 0.05, the symmetry assumption will lead to only a slightly over-estimated (about 2%) SIF, which will in turn lead to only a slightly underestimated propagation life (by about 3%). In summary, the wearnucleationpropagation approach gives better correlation with the test data for the lower slip amplitudes (partial slip), whereas the weartotal life approach gives better correlation for the higher

slip amplitudes (gross sliding). There are several works in the literature, where reasonable life predictions have been achieved considering only the propagation period, i.e. neglecting the nucleation period completely (e.g. [24]). However, the present work shows that the adoption of a propagation-only prediction model can lead to signicant inaccuracies. The balance between nucleation and propagation periods is dependent not only on the prevailing loading conditions, but also on the way in which the local stresses evolve with time. The bulk stress study shows how critical it is to consider the wear and propagation aspects of fretting modelling. Under gross sliding, it is found that the attenuation of stresses due to wear can result in crack arrest whereas the without wear analysis would fail to predict the potential benecial self arrest condition: the attenuation of stresses due to wear is sucient to stop the crack at a short length. Fig. 22 shows that initially the crack tip has sucient stress intensity to drive propagation, but the reduction in contact pressure due to the gross sliding condition causes the crack to arrest. The threshold envelope DKth,e depicted is bounded by the line s l K th l l0

K th;eff

21

This shows the value for DKe at which crack growth will occur at the current crack length, demonstrating a reduction in the threshold value at short crack lengths. The partial slip cases of Table 4 highlight that it would be incorrect to assume that neglecting wear just results in more conservative estimates for life when compared to an equivalent no-wear analysis.

Fig. 22. Comparison of with wear predicted DKe history with threshold envelope, for dapp,m = 8.7 lm gross sliding with rb = 100 MPa.

J.J. Madge et al. / International Journal of Fatigue 30 (2008) 15091528

1527

6. Conclusions The wearnucleationpropagation model presented in this paper is the rst attempt, to the authors knowledge, to address the interaction between crack nucleation and growth and wear eects in a nite element context. The paper has presented a study of the eects of wear on fretting fatigue crack propagation (and nucleation). The results show that the propagation of fretting fatigue cracks can be strongly aected by changes in the contact stress distributions which are caused by wear. The conclusions are as follows:  It is shown that the model captures the key eect of a minimum predicted fretting fatigue life with respect to a certain range of slip (or displacement) amplitude. This phenomenon has not been captured by other methodologies, excluding the previously-published wear-total life methodology of the authors. Furthermore, the inclusion of crack propagation eects with the wearnucleation approach leads to improved life prediction accuracy in the critical range of slip (minimum life), where the propagation life is predicted to form up to 20% of total life.  The evolution of contact stress under partial slip conditions increases the rate of propagation of cracks in the slip zone to levels similar to those at the edge of contact, thus capturing the experimentally-observed phenomenon of cracking within the slip zone, as well as at the contact edge. This is in contrast to models that neglect wear, which predict the highest propagation rates at the edge of contact.  For the loading conditions considered, the study of propagation in isolation does not give an accurate indication of total fatigue life. Nucleation can be a signicant component of total life.  A method has been suggested for deriving a set of shortscale (nucleation) SWT fatigue constants, by combining fracture mechanics with traditional long-scale SWT constants, to allow a more meaningful application of multiaxial fatigue damage models to problems involving high stress gradients, as found in contact fatigue.  It is shown that dierent loading combinations of the same initial geometry can produce markedly dierent relative contributions of propagation and nucleation to total life. However, it has been shown that the notion that the distinction between nucleation and propagation aspects of life can be neglected has merit under certain conditions.  The complex interaction between wear and fatigue has been highlighted by analyses using lower bulk fatigue loads: The wear resulting from gross sliding can cause self arrest in a case where a without wear analysis would predict crack propagation and component failure. Partial slip can result in a dormant crack propagating from the stick-slip boundary to cause failure. Critically, similar loading conditions used in a without

Fig. 23. Fretting map indicating the various wear induced phenomena and their relation to slip and bulk stress level.

The predicted eects of wear are complex. Depending on the slip distribution across the surface, and the substrate stress, not only is accelerated or retarded failure predicted, but dierent predictions with respect to nite or innite life are obtained. Fig. 23 shows the qualitative occurrence of these wear-induced phenomena in terms of the slip and rb variables on a fretting map. It is important to note that the present methodology employs several parameters that are purely empirical and have been obtained from fretting or other tests, such as stress- or strain-controlled fatigue tests, crack growth tests etc. Examples of these include coecient of friction, wear coecient, SWT constants, Paris constants and the short crack growth threshold. This is advantageous in the sense that this type of data is typically readily available, but nonetheless, it means that the present methodology is not a purely theoretical tool. It should be pointed out that the empirical parameters used in the model may be aected by normal load, material properties, surface roughness, surface contamination etc. Hence, due attention would need to be given to these considerations for dierent situations, since the values could be expected to be dierent to those employed here. The present methodology shows that material removal due to wear can have a signicant eect of crack initiation and propagation, so that in gross sliding, for example, the contact area increases due to material removal and hence the contact pressure reduces. However, as discussed in [39], for example, contact area can also increase due to junction growth, resulting from the combination of normal and tangential loading. This phenomenon is not included in the present methodology, but its absence may contribute to the discrepancies between the predicted and measured data and this aspect would need to be included in a more purely theoretical methodology.

1528

J.J. Madge et al. / International Journal of Fatigue 30 (2008) 15091528

wear analysis predicted that the crack tip loading remains below the threshold SIF, indicating innite life. Acknowledgements The authors wish to thank Rolls-Royce plc, Aerospace Group, for their nancial support of the research, which was carried out at the University Technology Centre in Gas Turbine Transmission Systems at the University of Nottingham. The views expressed in this paper are those of the authors and not necessarily those of Rolls-Royce plc, Aerospace Group. References
[1] Varenberg M, Etsion I, Halperin G. Slip index: a new unied approach to fretting. Tribol Lett 2004;17(3):56973. [2] Leen SB, Ratsimba CHH, McColl IR, Williams EJ, Hyde TR. An investigation of the fatigue and fretting performance of a representative aeroengine splined coupling. J Strain Anal Eng Des 2002;37(6):56585. [3] Szolwinski MP, Farris TN. Mechanics of fretting fatigue crack formation. Wear 1996;198:93107. [4] Smith KN, Watson P, Topper TH. A stressstrain function for the fatigue of metals. J Mater 1970;15:76778. [5] Fatemi A, Socie D. A critical plane approach to multiaxial fatigue damage including out of phase loading. Fatigue Fract Eng Mater Struct 1988;11(3):14965. [6] Araujo JA, Nowell D. The eect of rapidly varying contact stress elds on fretting fatigue. Int J Fatigue 2002;24:76375. [7] Sum WS, Williams EJ, Leen SB. Finite element, critical plane, fatigue life prediction of simple and complex contact congurations. Int J Fatigue 2005;27:40316. [8] Jin O, Mall S. Eects of slip on fretting behaviour: experiments and analyses. Wear 2004;256:67184. [9] Fouvry S, Duo P, Perruchaut P. A quantitative approach of Ti6Al 4V fretting damage: friction wear and crack nucleation. Wear 2004;257:91629. [10] Dang Van K, Papadopoulos IV. Multiaxial fatigue failure criterion: a new approach. In: Ritchie RO, Starke Jr EA, editors. Proceedings of the third international conference on fatigue and fatigue thresholds, fatigue 87. UK: EMAS Warley; 1987. p. 9971008. [11] Munoz S, Proudhon H, Dominguez J, Fouvry S. Prediction of crack extension under fretting wear loading conditions. Int J Fatigue 2006;28:176979. [12] Shkarayev S, Mall S. Computational modelling of shot-peening eects on crack propagation under fretting fatigue. J Strain Anal Eng Des 2001;38(6):495506. [13] Hattori T, Watanabe T. Fretting fatigue strength estimation considering the fretting wear process. Tribol Int 2006;39:11005. [14] Moobola R, Hills DA, Nowell D. Designing against fretting fatigue: crack self-arrest. J Strain Anal Eng Des 1997;33(1):1725. [15] Goh C-H, Wallace JM, Neu RW, McDowell DL. Polycrystal plasticity simulations in fretting fatigue. Int J Fatigue 2001;23:S42335.

[16] Madge JJ, McColl IR, Leen SB. A computational study on the wear induced evolution of fatigue damage parameters in fretting fatigue specimens. Proceedings of Fatigue 2006 (CD-ROM). GA, Atlanta: Elsevier; 2006. p. FT174. [17] Madge JJ, Leen SB, McColl IR, Shipway PH. Contact-evolution based prediction of fretting fatigue life: Eect of slip amplitude. Wear 2007;262(910):115970. [18] McColl IR, Ding J, Leen SB. Finite element simulation and experimental validation of fretting wear. Wear 2003;256:111427. [19] Jin O, Mall S. Eects of slip on fretting behaviour: experiments and analyses. Wear 2004;256:67184. [20] Ding J, Sum WS, Rajaratnam S, Leen SB, McColl IR, Williams EJ. Fretting fatigue predictions in a complex coupling. Int J Fatigue 2007;29:122944. [21] Chen WR, Keer L. Fatigue crack-growth in mixed-mode loading. J Eng Mater Technol Trans ASME 1991;113(2):2227. [22] El-Haddad MH, Smith KN, Topper TH. Prediction of non-propagating cracks. Eng Fract Mech 1979;11:57384. [23] Kitagawa H, Takahashi S. Applicability of fracture mechanics to very small cracks or the cracks in the early stages. Proceedings of the second international conference on mechanical behavior of materials. Metals Park, OH: American Society for Metals; 1976. p. 62731. [24] Navarro C, Munoz S, Dominguez J. Propagation in fretting fatigue from a surface defect. Tribol Int 2006;39:114957. [25] Sih GC. Strainenergydensity factor applied to mixed mode crack problems. Int J Fracture 1972;10(3):30521. [26] Sih GC. Mechanics and physics of energy density theory. Theor Appl Fract Mech 1985;4:15773. [27] Pawtucket HKS. Rhode Island. ABAQUS users and theory manuals version 6.6; 2006. [28] Sabelkin V, Mall S. Relative slip on contact surface under partial slip fretting fatigue condition. Strain 2006;42(1):1120. [29] Magaziner R, Jin O, Mall S. Slip regime explanation of observed size eects in fretting. Wear 2004;257:1907. [30] Wallace JM, Neu RW. Fretting fatigue crack nucleation in Ti6Al 4V. Fatigue Fract Eng Mater Struct 2003;26:199214. [31] Fadag HA, Mall S, Jain VK. A nite element analysis of fretting fatigue crack growth behaviour in Ti6Al4V. Eng Fract Mech 2007. doi:10.1016/j.engfracmech.2007.07.003. [32] Nicholas T, Huston A, John R, Olson S. A fracture mechanics methodology assessment for fretting fatigue. Int J Fatigue 2003;25:106977. [33] Dowling NE. Mechanical behaviour of materials: engineering methods for deformation, fracture and fatigue. 2nd ed. New Jersey: Prentice Hall; 1998. [34] Murakami Y. Stress intensity factors handbook, vol. 1. Pergamon Press; 1987. [35] Persson P-O, Strang G. A simple mesh generator in MATLAB. SIAM Rev 2004;46(2):32945. [36] Anderson TL. Fracture mechanics: fundamentals and applications. Florida: CRC Press; 1991. p. 714. [37] Rooke DP, Cartwright DJ. Compendium of stress intensity factors, H.M.S.O; 1976. [38] Lykins CD, Mall S, Jain VK. Combined experimental-numerical investigation of fretting fatigue crack initiation. Int J Fatigue 2001;23:70311. [39] Brizmer V, Kligerman Y, Etsion I. Elasticplastic spherical contact under combined normal and tangential loading. Tribol Lett 2007;25(1):6170.