Vous êtes sur la page 1sur 26

Gallego et al.

, 2007/05/11

Multiscale Computation of Fretting Wear at the Blade/Disk Interface

L. Gallego1, B. Fulleringer1, S. Deyber2, D. Nélias1


1
LaMCoS (UMR CNRS 5259), INSA Lyon, Villeurbanne, F69621, France, daniel.nelias@insa-lyon.fr
2
SNECMA, Villaroche, Moissy-Cramayel, F77556, France, stephane.deyber@snecma.fr

ABSTRACT

Dovetail joints between fan blades and the disk of turbine engines are subjected to fretting.

The objective of this research is to realize wear prediction by computational methods. The goal is

obviously the estimation of wear kinetics, but also to obtain worn surfaces, and permit the

manufacturer to realize complementary design analyses with worn surfaces. A wear law

developed for titanium alloy and based on the friction dissipated energy is used. A computational

method based on a three scale analysis is presented. The originality consists of coupling a semi-

analytical (SA) contact solver with the FE method for the structural behavior, allowing a fine

discretization of the contact zone. Contact computations are fast enough to realize cyclic wear

computations. Results for the blade/disk system are exhibited.

Keywords: fan, blade, disk, fretting, wear, stick-slip, contact computation.

1 INTRODUCTION

Fretting defines a contact situation submitted to small tangential displacements in

comparison to the contact dimensions. It means that a central part of the contacting surfaces are

always in contact during a fretting cycle. Alternating loading conditions are usually at the origin

of fretting. This phenomenon creates visible damages that lead to an accelerated ageing of the

functional surfaces. To predict service lives of mechanical joints, fretting should enter into

consideration. Two kinds of damages could be produced: i) Crack initiation and propagation.

Because of the fracture of the mechanical system that could happen, this situation is dangerous

1
Gallego et al., 2007/05/11

and should be avoided. ii) Wear of the contacting surfaces. If functional dimensions are spoiled,

vibrations (e.g. rolling bearing) or a loss of functionality could appear.

Dovetail joints between fan blades and the disk of turbine engines are subjected to fretting.

This is due to the centrifugal force (low frequency solicitation) that depends on the engine

regime and the aero-dynamical vibrations (high frequency solicitation) acting on the blade. It

causes a relative displacement of the contacting surfaces producing either mixed stick-slip or

gross slip. It is usually accepted that low frequency solicitations are the cause of the wear

process, while the high frequency solicitations would favor fretting fatigue by crack initiation

and propagation. In this specific application cracks are totally proscribed for safety reasons.

Conversely wear could be seen as beneficial since it permits to prevent the crack propagation by

removing surface cracks when initiated. However, for economical reasons, wear should be

controlled to limit maintenance operations.

The titanium alloy Ti-6Al-4V is commonly used for both disk and blades because of his

high mechanical properties and low density despite his poor tribological properties. Coatings

and/or solid lubricants are often used to control the friction coefficient all along the component

life while minimizing the wear. The objective of this research is to realize wear prediction by

computational methods. The goal is obviously the estimation of wear kinetics, but also to obtain

worn surfaces, and to realize complementary design analyses with worn surfaces.

To obtain realistic simulation results, two key points are required: i) an accurate wear law,

and ii) a fast and robust three-dimensional contact computation tool.

The tribological analysis of a contact can be understood through the energetic balance.

Friction is a dissipative phenomenon, and a part of this friction dissipated energy is consumed by

2
Gallego et al., 2007/05/11

different processes (material transformations, physical-chemical processes, third body formation)

that conduct to wear [1]. To quantify the wear of a sliding contact, Archard [2] published in 1953

P
his famous equation: V = α ARCHARD ⋅ s ⋅ where αARCHARD is the wear coefficient. The worn
K

volume V is function of the sliding distance, s, the normal applied load, P, and the hardness of

the softer material, K. The friction coefficient does not appear in this equation. However it is

often observed that the greater the friction coefficient, the larger the wear. In addition, the

friction dissipated energy is equal to the work of the tangential load. For these reasons Fouvry

and Paulin [3-5] developed a modified Archard’s wear law, based on the friction dissipated

energy. This wear law integrates the friction coefficient and the tangential load. This law has

been established for fretting contact and its reliability has been proved over a relatively wide

range of normal loads and sliding amplitudes in the case of cylinder on flat fretting tests. The

wear volume is associated to the friction dissipated energy by a wear coefficient. The formula is:

V = α ∑ Ed (1)
N

Fouvry and co-workers have also given an extension in view of introducing the adhesion

phenomena that affects ductile materials such as titanium or aluminum alloys. They have

observed that the wear coefficient increases proportionally to the fretting tangential

displacements δ0. Godet and co-workers [6,7] have shown that wear is a consequence of debris

formation and debris ejection. In a very simple manner Ed can be linked to the formation of

debris and δ0 to the debris ejection. Therefore the wear coefficient is balanced by δ0, and the

wear law is written:

δ0
V = α ref
δ 0 ref
∑ Ed
N
(2)

3
Gallego et al., 2007/05/11

where ref refers to a reference fretting test in term of sliding amplitude, Ed is the energy

dissipated by friction and Et is the total energy. This wear law is modified to a local writing. δ0

which can be interpreted as the slip amplitude at the centre of the contact is replaced by the true

slip amplitude at each point of the contact area. Finally, the wear law consists in a simple

geometrical modification of the contact surfaces defined by a wear depth:

 α ref   
∆h =  4 ∑  ∑ ∆s t ⋅ ∑ q t ⋅ ∆s t  (3)
 δ 0 ref  N 1 cycle
   1 cycle 

Finite element (FE) models have been developed by many authors [8-10] for wear

computations. The methodology consists in computing one fretting cycle load history, post

processing the contact data fields, obtaining a wear depth field, updating the mesh by moving the

node at each contact interface, and restarting this procedure iteratively until a final number of

cycles. For three dimensional fan blade models, where a fine mesh is required for the contact

interfaces, this leads to high computing costs, incompatible with an industrial design process.

The current paper presents a different methodology based on a three scale analysis. The first

scale is the micro-scale, where the wear depth increment is calculated depending on the contact

solicitations, i.e. contact pressure, shear stress and slip amplitude. These solicitations are derived

from a contact analysis at the meso-scale, starting from the knowledge of the forces and the

moments at a given point of the contact. The latter being the result of a FE analysis of the

complete disk and blade structure. The originality consists of coupling a semi-analytical (SA)

contact solver with the FE method for the structural behavior, allowing a fine discretization of

the contact zone. The SA code has been already used in previous studies to predict fretting wear

in the case of academic fretting tests such as the cylinder on plane fretting test [11,12]. It is worth

4
Gallego et al., 2007/05/11

noting that the method, which is here limited to elastic behavior, could be quite easily extended

to elastic-plastic [13,14], thermal elastic-plastic [15] or layered [16] contact.

2 CONTACT MODEL OUTLINE

2.1 The finite element model

A finite element model of the fan blade and disk structure is built. One low frequency

fretting cycle is simulated. The aero dynamical loading known in advance is forced and an

increasing and then decreasing centrifugal loading is imposed (cf. fig.1). The model is elastic,

static, and geometrically non-linear. For each interface contact output data are memorized all

along the loading path: pressures, shear tractions and slips. The Abaqus standard solver has been

used with 22 increments to describe the loading path. For instance the behavior of the intrados

contact is outlined. Both full slip and stick/slip periods coexist during a loading cycle. The

contact solver should be able to describe these two cases.

2.2 From the finite element model to the semi-analytical model

The first step to set up the SA analytical contact model is to define the local contact

geometry for each interface. A local system of coordinate (O,x,y,z) is chosen with origin at the

center of each contact interface (cf. fig.2). The FEM pressures and shears are summed to obtain

the normal load P and the tangential loads (Qx, Qy) for each dovetail face. The moments Mx, My

and Mz are computed in the same way around the origin of the local coordinate of each contact.

Additional terms are calculated from the FE results: the friction dissipated energy along each

direction. These two terms will intervene in the full slip contact computation case.

2.3 The semi-analytical model

The initial contact geometry is defined on a regular grid, with uniform pitch on each

direction. This grid is the potential contact area (Γp) and should be greater than the contact area

5
Gallego et al., 2007/05/11

(Γc). The undeformed gap distance h(x,y) is the difference between the two surfaces of equation

z=surf1(x,y) and z=surf2(x,y). 1 refers to the disk and 2 refers to the blade. These surfaces consist

of flat plane and round edges. The aim of the SA contact solver is to solve the static contact

equilibrium, i.e. to solve the normal and tangential contact, and for each time increment. The

elastic half space assumption is done. Therefore Boussinesq and Cerruti potentials [17,18]

integrated to a normal and tangential uniformly loaded rectangular area give the influence

coefficients K that link the elastic deflections in the normal (ūz) and tangential directions

 ux   qx 
( uτ =   ) to the pressures (p) and shear tractions ( q =   ). Let t the superscript that refers to
uy   qy 

the time, i.e. the increment number of the previous FEM computation. Between each increment,

the tangential elastic deflection is defined by uτt = uτt −1 + ∆ uτt . The following discrete

convolutions are obtained:

 t
(qx t
) (
qy t
)
∆u x ij = ∑ K x ijkl ⋅ q x kl − q x kl + ∑ K x ijkl ⋅ q y kl − q y kl + ∑ K x ijkl ⋅ p kl − p kl
t −1 t −1
( p t t −1
)
 (i , j )∈Γc (i , j )∈Γc (i , j )∈Γc
t
(qx t
) (
qy
)
∆u y ij = ∑ K y ijkl ⋅ q x kl − q x kl + ∑ K y ijkl ⋅ q y kl − q y kl + ∑ K y ijkl ⋅ p kl − p kl
t −1 t t −1
( p t
)
t −1

 (i , j )∈Γc (i , j )∈Γc (i , j )∈Γc


u z ij = ∑ K z ijkl ⋅ q x kl + ∑ K z ijkl ⋅ q y kl + ∑ K z ijkl ⋅ p kl
t qx t qy t p t

 (i , j )∈Γc (i , j )∈Γc (i , j )∈Γc


(4)

The influence coefficients are function of E1, E2, ν1, ν2. When E1=E2 and ν1=ν2, it follows
that:

 t
(qx t t −1
) qy t
(
∆u x ij = ∑ K x ijkl ⋅ q x kl − q x kl + ∑ K x ijkl ⋅ q y kl − q y kl
t −1
)
 (i , j )∈Γc (i , j )∈Γc
t
(qx t
) qy t
(
∆u y ij = ∑ K y ijkl ⋅ q x kl − q x kl + ∑ K y ijkl ⋅ q y kl − q y kl
t −1 t −1
) (5)
 (i , j )∈Γc (i , j )∈Γc
u z ij = ∑ K z ijkl ⋅ pkl
t p t

 (i , j )∈Γc

6
Gallego et al., 2007/05/11

The normal and the tangential contact problems are therefore uncoupled. The normal

contact problem consists to find the pressure distribution that permits to obtain a gap nil (eq. 6.b)

in the contact area (eq. 6.a) and positive (eq. 6.d) outside the contact area (eq. 6.c). The

equilibrium in terms of normal load P and flexion moment Mx and My has to be checked also.

The unknowns included in the gap are the elastic deflections, a rigid body displacement (δz) and

two small rigid body rotations (φx, φy). For the tangential problem, between each increment, a

finite shift is assumed. The shear stress is opposite to the slip direction and equal to the product

of the friction coefficient by the pressure (eq.7.a) in the slip area Γsl. In the stick area Γst the

absolute shear stress is lower than the product of the friction coefficient by the pressure (eq.7.c).

The slip (eq. 7.b and eq. 7.d) is the difference between the tangential elastic displacements and a

tangential rigid body displacement δτ and the displacements resulting from a small rigid body

rotation φz. The equilibrium in terms of tangential load Q and torsional moment Mz has to be

checked. When computing increment with a gross slip behavior, this system of equations does

not remain available and is replaced by eq.8. Indeed, the shear stress amplitudes are known in

advance because all the contact area is slipping. The equilibrium equations are removed. To

obtain the rigid body displacement, two new equations are needed. The friction dissipated energy

in both directions is computed from the FEM model and is added as new inputs of the contact

solver. These two equations (eq. 8.a, eq. 8.b) are necessary to estimate the tangential rigid body

displacement. The equilibrium moment is here not taken into account, indeed when full slip is

reached the torsional rigid body angle contribution is negligible compared to the tangential body

displacement. Because contact problems can be presented as a problem of minimization of the

complementary energy [19, 20], numerical technique for constrained optimization should be

used. The numerical techniques used to solve the normal contact equations are a conjugate

gradient method presented elsewhere in [21], and the Discrete Convolutions and Fast Fourier

7
Gallego et al., 2007/05/11

Transforms technique (DC-FFT), introduced by Liu et al [22] to reduce the computation time of

the discrete convolutions (eq.4) and that requires to extend the grid only by two. The conjugate

gradient is here adapted to take into account the flexion moments. A similar procedure is set up

to compute the tangential problem.

 pijt > 0, ∀(i,j ) ∈ Γc (a )


 t
hij + u z ij − δ z − φ x ⋅ yij + φ y ⋅ xij = 0, ∀(i,j ) ∈ Γc (b )
t

 pijt = 0, ∀(i,j ) ∉ Γc (c )
 t
hij + u z ij − δ z − φ x ⋅ yij + φ y ⋅ xij > 0, ∀(i,j ) ∉ Γc (d )
t t

 pt ⋅ S = Pt (e )
∑ Γp
ij
(6)

∑ pij ⋅ yij ⋅ S = M x
t t
(f )
 Γp
∑ − pij ⋅ xij ⋅ S = M y
t t
(g )
 Γp

q tij = µ ⋅ pijt ⋅ ∆s tij ∆s tij , ∀(i, j ) ∈ Γsl (a )



t
 t yij 
∆ u − ∆ δ t
− ∆ φ 
z
 = ∆s tij ≠ 0, ∀(i, j ) ∈ Γsl (b )
 ij 
 t  − xij 
 q ij < µ ⋅ pij , ∀(i, j ) ∈ Γst (c )
t

∆ u t − ∆δ t = 0, ∀(i, j ) ∈ Γst (d )
 ij t x
(7)
∑ q ij ⋅ S = Q
t
(e)
 Γc t
(
∑ q yij ⋅ xij − q xijt
)
⋅ yij ⋅ S = M zt (f )
 Γc
Γsl ∪ Γst = Γc (g )

8
Gallego et al., 2007/05/11

q tij = µ ⋅ pijt ⋅ ∆s tij ∆s tij , ∀(i, j ) ∈ Γc (a )



t
 t yij 
∆ u − ∆ δ t
− ∆ φ 
z
 = ∆s tij ≠ 0, ∀(i, j ) ∈ Γc (b )
 ij 
  − xij 
∑ − q xijt
⋅ ∆s xij
t
⋅ S = Ed xt (c )
 Γc (8)
 − q t ⋅ ∆s t ⋅ S = Ed t
∑ yij yij y (d )
Γc
Γsl = Γc
 (e )

2.4 Validation of the semi-analytical model

The normal contact solver and the partial slip contact solver are first validated by

comparing numerical results to the analytical solution of a sphere in contact with a flat with

identical elastic properties.. The pressure distribution is given by Hertz [17], whereas Mindlin

[23] provided the solution of the tangential problem when the shear stress is controlled by the

Coulomb friction law. Both analytical and numerical solutions exhibit a slip annulus (cf. fig.4) at

the boundary of the contact area. A fretting loop can also be realized to compare the ability of the

code to apprehend several loading increments (cf. fig.5 and fig.6). These results have already

been presented in [12]. The titanium elastic constants are used: E=115GPa, ν=0.29. The

tangential and normal load paths are equivalent to the ones usually applied to fretting tests. The

normal load is kept constant (400 N) while an alternative tangential load is imposed in the x-

direction. The maximal tangential load (210N) is lower than the limiting load due to Coulomb’s

friction law (µ=0.7). The radius of the sphere is 10 mm.

A solution to check the aptitude of the code to solve contact problem under torsional

moment is the solution provided by Lubkin [24] for a sphere subjected to a normal load and

torsional moment. Similarly to the normal and tangential problem one may observe a slip

annulus at the boundary of the contact area, the normal pressure being locally sufficiently low to

9
Gallego et al., 2007/05/11

allow surfaces to slip. The magnitude of the surface traction is given in fig. 7 when a normal load

of 400N is kept constant and a torsional moment of 10 N.mm applied. A very good agreement

between numerical and analytical results is found, as for the torsional angle. It should be

emphasized that the same shear distribution is obtained in terms of amplitude, sign and directions

(i.e. the radial component is nil).

3 WEAR MODELING

Two methods are now proposed to perform wear computations. The first one (cf. fig.8)

consists of realizing the normal contact computation with the contact solver and assuming the

slips output of the FE model. The wear is computed using eq.3 and replacing the shear with the

product of the normal load by the friction coefficient. In the second method (cf. fig.9), the

contact solver is used to solve both the normal and tangential contact problems. The first method

is the faster in term of computing time since it requires only 30s on a regular laptop to perform

wear computation during one cycle (i.e. 22 increments). In fact the contact computation is

limited to the normal contact problem since slips are provided by the FEM model. In the slip

zone the wear depth is calculated using the pressures from the SA code multiplied by the friction

coefficient to obtain surface traction. The second method is more costly in term of computing

time since it requires solving both the normal and tangential problem, which are moreover

coupled when the elastic properties are different for the contacting surfaces. It corresponds to

190s on the same laptop to perform the wear computation for 22 increments.

The great interest of such a method is that the wear process consists simply in readjusting

the heights of the surfaces on the grid after each cycle for further FE analysis. It permits to avoid

the delicate remeshing stage of the FEM. It is also useless to update the geometry of the

contacting surfaces after each fretting cycle, since the wear rate is usually very small. Therefore

10
Gallego et al., 2007/05/11

the contact analysis, which differs only if the geometry is changed significantly between two

calculations, can be performed only each ∆N cycle increment, the wear increment being then

multiplied by ∆N. It results in an increase of the computing speed by the same factor. The next

cycle to be simulated becomes N = N + ∆N. The acceleration term can be imposed by the

engineer based on previous experience or automatically computed through several criterions, for

instance a maximum wear depth, a wear volume, etc… Nevertheless this acceleration term must

be kept small enough to avoid surface discontinuities or creation of steps on the new surfaces.

Finally one last criterion should be defined to stop the wear computation, e.g. a maximal number

of cycles, a maximum wear depth, etc …

4 APPLICATION TO A DISK-BLADE CONTACT

A mission loading is applied to compute the extrados and intrados contacts over one low

frequency cycle. Both contacts are analyzed simultaneously but results presented here are limited

to the intrados for illustration purpose. In the present analysis the normal load and the flexion

moments are taken into account, but slips are provided by the FE analysis. Note that a biaxial

tangential load and the torsional moment are taken into account in the contact analysis by the SA

method. From the FEM results, it appears that the intrados contact is in gross slip until increment

8 during the increase of the centrifugal load. The tangential load is not used as an input of the SA

code in this situation, since it does not permit to forecast the final slip amplitude. An equilibrium

in terms of friction dissipated energy is used instead. It will be verified that the tangential load

equilibrium is realized naturally.

The evolution of both components of the tangential load is plotted in fig. 10. Values from

the FEM and from the SA code are found to be in good agreement. Another point of interest is

the evolution of the torsional moment during the full slip stage since contrary to the stick-slip

11
Gallego et al., 2007/05/11

stages, the torsional moment equilibrium is not enforced anymore. Nevertheless the FEM results

and the SA results agree very well as shown in fig. 11 – including during the 8 first increments

corresponding to gross slip. It validates a posteriori the assumption that the contribution of a

rotational body displacement is negligible compared to the translating tangential body

displacement.

Complementary results on the pressure, shear traction and slip distributions are given and

compared to the results of the FE model. These results are given at 3 typical increments.

Increment 8 is the increment where the maximum normal load is reached and the contact is fully

sliding. Increment 12 is an increment with a small tangential load and a stick-slip contact

behavior. Increment 15 still corresponds to stick-slip contact behavior but with a reversed

tangential load.

Figure 12 displays the profiles of the pressure distribution at increment 8 along the axes

y=0 and x=0. Results are coherent between both computation but the efficiency of the contact

code permits to use fine mesh and to observe details that can not be apparent with the

discretization of the FE model. The main difference is that the fine mesh of the SA model

permits to detect pressure peaks that are not predicted by the FE analysis which uses a coarser

mesh. The consequence of the mesh refinement can be outlined achieving a computation with the

SA code with a geometry defined on a grid with a refinement close to the FE model.

Corresponding pressure profiles are plotted in fig. 13. A coarser grid size for the SA contact

solver leads to results closer to those of the FE analysis. It is clear here that the contact mesh is

an important issue for the contact behavior and therefore the wear prediction.

Shear results provided by the SA code and the FE analysis are now presented in fig. 14 for

increments 8, 12 and 15. All components are given along the axes y=0 and x=0. At increment 8,

12
Gallego et al., 2007/05/11

the intrados contact is submitted to full slip. Shear peaks appear thanks to the fine mesh of the

SA model. Between the FE model and the biaxial SA model, a very localized sign difference

exists for the y component. At increment 12 the intrados contact is submitted to partial slip. The

SA model gives a solution close to the FE results. At increment 15 the intrados contact is

submitted to partial slip. Sign differences are noticed locally for the shear peaks. To conclude on

the traction distributions obtained, results are coherent with the FE ones, but the comparison

seems to be limited because of the quality of the FE analysis.

Figure 15 displays the profiles of the slip components sx and sy at increment 8. The mean

values of the slip in each direction are equivalent between SA results and FE results. However

the SA slip profiles are much more constant. It indicates that the outcome of the elastic

displacements is negligible compared to the rigid tangential displacement. Figure 15 presents

also the profiles of the slip components sx and sy at increment 12. The tangential load is then too

small to permit slipping within the contact area. Localized slip zones are found at increment 15

at the border of the intrados contact interface. However some differences are found between the

SA model and the FE analysis. First for the SA model localized slip appears on each corner. The

central part of the slip is common for both model except that the maximal value for the SA

model is about 2/3 the maximal value of the FE model.

5 WEAR RESULTS

The methodology permits to study the evolution of the wear and to determine wear kinetics

for each faces of the dovetail joint for example in terms of wear volume (cf. fig.16) or wear

depth (cf. fig.17). Because wear leads to an accommodation of the surfaces to prevent wear, the

new profiles will erase the initial pressure peaks and the pressure distribution will become

flattened. It results in a linear behavior of the wear kinetics after a given number of cycles

13
Gallego et al., 2007/05/11

(running-in period). From this linear behavior, an analytical estimation of the wear kinetics can

be made without any additional numerical computation. Finally fatigue life computation can be

realized in a second step and will provide more realistic results when a worn geometry is used

rather than when based on the initial unworn geometry.

An important and restrictive hypothesis that has been made in this investigation is that the

loads and moments applied to the contacts do not vary despite wear; they are still the result of the

initial FE analysis. However, it can be shown that all these output data remain stable despite the

wear for this specific application. For other applications this statement should be finely shaded. a

few additional (intermediate) FE computations should be needed to readjust the components of

the external loads and moments applied to each contact.

6 CONCLUSION

An alternative method to the FE method is exhibited to realize wear computations on

dovetail joints. However a FE model is necessary to solve the structural problem. The SA code is

used as a zoom on each contact area, and its quickness permits to compute easily cyclic wear.

Time computations are reduced using such a method compared to full FE wear prediction.

Contact output fields such as pressure, shear and slip distributions are found when using the SA

method. Comparison with the FE model is coherent but, the mesh size limits the comparison and

origins in the slip profiles differences have to be explored.

REFERENCES

[1] Kapsa P, Cartier M. Usure des Contacts Mécaniques – Problématiques et Définitions.


Techniques de l’Ingénieur 2001;BM 5 067:1-14.

[2] Archard JF. Contact and Rubbing of Flat Surfaces. J. Appl. Phys. 1953;24:981–988.

[3] Fouvry S, Duó P, Perruchaut Ph. A Quantitative Approach of Ti-6Al-4V Fretting Damage:
Friction, Wear and Crack Nucleation,. Wear 2004;257:916-929.

14
Gallego et al., 2007/05/11

[4] Paulin C, Fouvry S, Deyber S. Wear Kinetics of Ti–6Al–4V under Constant and Variable
Fretting Sliding Conditions. Wear 2005;259:292-299.

[5] Paulin C. Etude de l'Endommagement du Contact Multicouche Aube/Disque Sous


Chargement de Fretting : Impact des Sollicitations Variables et de la Dimension du
Contact. Ph.D. thesis, Ecole Centrale de Lyon, France, 2006.

[6] Godet M. The Third Body Approach, a Mechanical Point of View of Wear. Wear
1984;100:437-452.

[7] Berthier Y, Vincent L, Godet M. Velocity Accommodation in Fretting. Wear 1988;125:25-


38.

[8] Põdra P, Andersson S. Simulating Sliding Wear with Finite Element Method. Tribol. Int.
1999;32: 71–81.

[9] McColl IR, Ding J, Leen SB. Finite Element Simulation and Experimental Validation of
Fretting Wear. Wear 2004;256:1114–1127.

[10] Ding J, Leen SB, McColl IR. The Effect of Slip Regime on Fretting Wear-Induced Stress
Evolution. Int. J. Fatigue 2004;26:521–531.

[11] Gallego L, Nélias D, Jacq C. A Comprehensive Method to Predict Wear and to Define the
Optimum Geometry of Fretting Surfaces. ASME J. Tribol. 2006;128(3):476-485.

[12] Gallego L, Nélias D. Modeling of Fretting Wear under Gross Slip and Partial Slip
Conditions. ASME J. Tribol. 2007;129(3,in press).

[13] Nélias D, Boucly V, Brunet M. Elastic-Plastic Contact Between Rough Surfaces: Proposal
for a Wear or Running-in Model. ASME J. Tribol. 2006;128:236-244.

[14] Antaluca E, Nélias D, Cretu S. A Three-Dimensional Friction Model for Elastic-Plastic


Contact with Tangential Loading — Application to Dented Surfaces. Proceedings of 2004
ASME/STLE International Joint Tribology Conference, 2004, New York: ASME, Paper
No. TRIB2004-64331.

[15] Boucly V, Nélias D, Liu SB, Wang QJ, Keer LM. Contact Analyses for Bodies with
Frictional Heating and Plastic Behavior. ASME J. Tribol. 2005;127:355-364.

[16] Polonsky IA, Keer LM. A Fast and Accurate Method for Numerical Analysis of Elastic
Layered Contacts. ASME J. Tribol. 2000;122:30-35.

[17] Johnson KL. Contact Mechanics. London: Cambridge University Press, 1985.

[18] Love AEH. A Treatise on the Mathematical Theory of Elasticity, 4th ed. Cambridge:
Cambridge University Press, 1927.

[19] Kalker JJ. Three Dimensional Elastic Bodies in Rolling Contact. Dordrecht: Kluwer
Academic Publishers, 1990.

15
Gallego et al., 2007/05/11

[20] Kalker JJ. Variational Principles of Contact Elastostatics. J. Inst. Maths Applics
1977;20:199-219.

[21] Polonsky IA, Keer LM. A Numerical Method for Solving Rough Contact Problems Based
on the Multi-Level Multi-Summation and Conjugate Gradient Techniques.
Wear1999;231:206–219.

[22] Liu S, Wang Q, Liu G. A Versatile Method of Discrete Convolution and FFT (DC-FFT)
for Contact Analyses. Wear 2000;243:101–111.

[23] Mindlin RD. Compliance of Elastic Bodies in Contact. ASME J. Appl. Mech.
1949;16:259-268.

[24] Lubkin JL. Torsion of Elastic Spheres in Contact. ASME J. Appl. Mech. 1951;18:183.

Figure 1: Centrifugal load path

Figure 2 : Local coordinate system on the intrados dovetail joint face

16
Gallego et al., 2007/05/11

Figure 3: Slip behavior of the intrados contact

Figure 4 : Distribution of slip when an elastic sphere is pressed against an elastic flat during a linear fretting
cycle

17
Gallego et al., 2007/05/11

Figure 5: Distribution of surface traction when an elastic sphere is pressed against an elastic flat during a
linear fretting cycle

Figure 6: Fretting loop of an elastic sphere pressed against an elastic flat with an alternating tangential load

18
Gallego et al., 2007/05/11

Figure 7: Distribution of surface traction when an elastic sphere is pressed against an elastic flat during a
torsional fretting cycle

19
Gallego et al., 2007/05/11

Figure 8: Wear computation method - SA code used to solve only the normal contact problem

20
Gallego et al., 2007/05/11

Figure 9: Wear computation method - SA code used to solve the normal and tangential contact problem

21
Gallego et al., 2007/05/11

Figure 8: Evolution of the contact load components

Figure 9: Evolution of the contact moment components

22
Gallego et al., 2007/05/11

Figure 10: Comparison of pressure distribution between the FE and SA models (mesh 128x256 for the SA
code)

Figure 11: Comparison of pressure distribution between the FE and SA models with a coarser mesh (mesh
20x34 for the SA code)

23
Gallego et al., 2007/05/11

Figure 12: Comparison of shear distributions between the FE and SA models

24
Gallego et al., 2007/05/11

Figure 13: Comparison of slip distributions between the FE and SA models

25
Gallego et al., 2007/05/11

Figure 14: Evolution of the maximal wear depth

Figure 15: Evolution of the wear volume

26

Vous aimerez peut-être aussi