Vous êtes sur la page 1sur 12

International Journal of Fatigue 25 (2003) 547558 www.elsevier.

com/locate/ijfatigue

Rolling contact fatigue analysis of rails inculding numerical simulations of the rail manufacturing process and repeated wheelrail contact loads
Jonas W. Ringsberg a,, Torbjo rn Lindba ck b
a

Chalmers University of Technology, Department of Applied Mechanics, SE-412 96 Go teborg, Sweden b Lulea University of Technology, Division of Computer Aided Design, SE-971 87 Lulea , Sweden Received 21 June 2002; received in revised form 30 September 2002; accepted 3 October 2002

Abstract The present work is an investigation on how an initially introduced residual stress-state affects the service life of a rail, i.e. the time to fatigue crack initiation. The nite element (FE) method was used to make two-dimensional thermo-mechanical analyses of the rail cooling and roller straightening processes. The results became the initial conditions in a three-dimensional elastic-plastic rail model; the model is part of an FE tool developed for rolling contact fatigue (RCF) analysis of rails. The results from this tool were analysed for fatigue, for eight wheel passages, according to a method which incorporates a critical plane approach that evaluates fatigue damage on a cycle-by-cycle basis. A heavy-haul (30 tonne) train trafc situation on the Iron-ore Line in Sweden was studied with respect to subsurface fatigue crack initiation in straight track. Three examples using the rail model in the FE tool were assessed: (a) an initially stress-free rail, (b) a measured residual stress eld in a newly manufactured rail, and (c) a calculated residual stress eld from the cooling and roller straightening analyses. The results from the thermo-mechanical FE analyses of the rail manufacturing process showed tensile residual stresses in the longitudinal direction of the rail; this was validated with experimental measurements on newly manufactured rails. The FE tool and fatigue calculations revealed only small differences in results for the three examples. It was concluded that, because of the very high axle load in the present trafc situation, the local wheel-rail contact loads governed the fatigue life to crack initiation. Additional FE tool calculations were made to show the axle load at which rail manufacturing stresses reduce the fatigue life to crack initiation. 2003 Elsevier Science Ltd. All rights reserved.
Keywords: Rolling contact fatigue; FE simulations; Critical plane approach; Residual stresses; Rail fatigue; Rail manufacturing

1. Introduction Repeated railway wheelrail rolling contacts cause a nonproportional multiaxial stressstrain response in the rail. This causes fatigue damage to the rail, which leads to the initiation of cracks that originate on either the surface or subsurface of the rail. The site of the greatest fatigue damage is governed by factors including the magnitude of the traction forces, the axle load, and the residual stress eld in the rail. In addition, future
Corresponding author. Chalmers University of Technology, Department of Applied Mechanics, SE-412 96 Go teborg, Sweden. Tel.: +46-31-772-1504; fax: +46-31-772-3827. E-mail address: jonas.ringsberg@me.chalmers.se (J.W. Ringsberg).

demands on train trafc will involve heavier axle loads, higher train speeds and increased trafc density. Each of these demands increases rail rolling contact fatigue (RCF) damage, raises maintenance costs and affects train trafc safety. To prevent the future demands from resulting in catastrophic rail failures, it is urgent to conduct investigations that identify and suggest improvements for the rails available today. Ringsberg [1,2] developed a nite element (FE) tool for RCF analysis of railway rails; see Section 3 for description of the FE tool. It comprises two coupled FE models, one for the track and one for the rail, which together incorporate the global track response and the local wheel-rail contact loads in the analysis. The tool was employed to analyse suburban train trafc and heavy-haul train trafc situations with respect to RCF

0142-1123/03/$ - see front matter 2003 Elsevier Science Ltd. All rights reserved. doi:10.1016/S0142-1123(02)00147-0

548

J.W. Ringsberg, T. Lindba ck / International Journal of Fatigue 25 (2003) 547558

crack initiation. The fatigue calculations were made using a method proposed for life prediction of RCF crack initiation [3,4]. Furthermore, the FE tool assumed that the rail was stress-free at the start of the analysis, i.e. residual stresses caused by the rail manufacturing process were not incorporated. Schleinzer [5], Lindba ck [6], Webster [7], and Urashima [8] have done experimental work and made numerical simulations of rail manufacturing processes. They showed that the rail heads of newly manufactured rails often had a residual stress-state of tensile stresses in the longitudinal direction near the running surface. A residual stress-state of this type can reduce the time to fatigue crack initiation. The objective of the present investigation was to examine whether the residual stresses caused by the rail manufacturing process inuence the fatigue life to crack initiation during multiaxial fatigue loading conditions. This was accomplished by further development and improvement of the work done by Lindba ck [6] and Ringsberg [14]. The work was divided into three parts: computational modelling of the rail manufacturing process; FE tool analyses of wheelrail rollingsliding contact loads subjected to rails during service conditions; presentation and utilisation of a method proposed for fatigue life prediction of RCF crack initiation. Inspections of damaged pearlitic steel grade 1100 rails, along the heavy-haul (30 tonne axle load) Ironore Line in Sweden, show early subsurface fatigue crack initiation in straight track. The hypothesis was that the residual stress-state caused by rail manufacturing affected the time to fatigue crack initiation and the position for fatigue failure. The train trafc situation at the Ironore Line was therefore chosen for study in this investigation. Three FE tool analyses were made using different initial stress-state conditions of a rail: (a) stress-free rail, (b) measured residual stress eld of newly manufactured rails, and (c) calculated residual stress eld from a rail manufacturing simulation. Each of the examples was analysed for fatigue. Hence, the inuence of the initial stress-state of a rail, introduced during rail manufacture, on the fatigue life to crack initiation could be determined.

sequences through machinery that forms it to the correct rail prole. The nished hot-rolled rail prole has a temperature of approx. 900 C just before it enters a cooling bed. The hot-rolled rail is pushed onto the cooling bed where it is laid on its side to cool almost to room temperature, see Fig. 1. When the rail is pushed onto the cooling bed, it is curved around an axis normal to the bed (the y-axis) to compensate for the change in curvature it will undergo during cooling. The curvature change occurs because different parts of the rail are cooled at varying rates depending on rail shape and thickness. When the rail has cooled to room temperature, it enters a roller straightening process. The rail is straightened vertically by running it through nine rollers. The position of the rollers is designed to alternately bend the rail beam up and down, so that a straight rail passes the ninth roller. The rail manufacturing process was simulated with an in-house thermo-mechanical FE code. A two-dimensional rail FE model based on a generalised plane deformation formulation was used [9]. The out-of-plane strain (i.e. the strain in the longitudinal direction of the rail) was assumed to have the linear variation ex = b1 + b2y + b3z, i.e. three unknowns were added. In contrast to a three-dimensional model, information on the out-ofplane shear strains is lost, i.e. exy = exz = 0. The rail cross-section was divided into 576 four-node quadrilateral elements (see Fig. 2) with bilinear shape functions. 2.1. Simulation of cooling The cooling simulation began when the hot-rolled rail, which has a temperature of 900 C, was pushed onto the cooling bed. At this temperature, the yield limit of the material is low and its stress relaxation is fast. The magnitude of the residual stresses in the rail is, at this point, negligible when compared with the magnitude of the resultant residual stresses after cooling and roller straightening. The rail was assumed xed with respect to bending around its symmetry line in the z-direction (see Fig. 1). This assumption was based on the fact that the weight of the rail prevents it from moving out of the cooling bed plane. During cooling, heat was emitted from the rail due to heat transfer to the cooling bed, convection, and radiation. The heat transfer to the cooling bed, which acts as a heat sink that affects the cooling rate, was modelled by two extra elements at the contact points between the rail and the cooling bed. The two elements were given a heat conduction coefcient of low value, corresponding to the heat transfer in the contact. The element nodes in these two extra elements not connected to the rail were given a temperature equivalent to that of the cooling bed. Convection and the magnitude of the convective heat transfer coefcient are highly dependent on the properties and state of the cooling uid (in this case

2. Rail manufacturing simulation The rail manufacturing process comprises three steps: hot-rolling; cooling; and roller straightening. In each of the three steps, residual stresses are introduced in the rail beam. The resultant stress eld is complex to estimate in advance, as it is a result of successive steps each of which affects the stress-state. The present investigation integrates the simulation of the cooling and roller straightening steps. Hot-rolling begins with an iron bloom that is heated to over 1000 C. The bloom undergoes several hot-rolling

J.W. Ringsberg, T. Lindba ck / International Journal of Fatigue 25 (2003) 547558

549

Fig. 1.

Hot-rolled rail prole pushed onto cooling bed; compensation for curvature changes is followed by roller straightening.

Fig. 3. The thermal dilatation (et) and the yield stress (sy) vs. the temperature (T).

Fig. 2. FE mesh of rail cross-section in the cooling and roller straightening FE simulations.

dependent mechanical properties for the pearlitic steel grade 900A [11], see Figs. 3, 4 and 5 for the temperature-dependent mechanical properties. The material behaviour was modelled using a constitutive material model for linear kinematic hardening.

free air). According to Kreith and Bohn [10], the convective heat transfer coefcient for convection between steel and free air varies between 6 and 30 W/m2 C. In the present investigation, it was set to 12 W/m2 C to t cooling rates measured at the rail manufacturing plant. The heat loss due to radiation was calculated using the Stefan-Boltzmann law with a unit emission factor. The rail was cooled for 8 hours, after which it had reached room temperature. A staggered step-approach was used in the cooling simulation. For each time step, the thermal eld was calculated using an iterative procedure, followed by iterations to nd a solution to the mechanical eld. The cooling rate during the simulation was approximately 0.5 C/s, and the transformation from austenite to pearlite for the current steel takes place between 600 C and 650 C. The thermal and mechanical elds were coupled in the FE simulation by thermal strains and temperature-

Fig. 4. The hardening modulus (H), the elastic modulus (E), and the Poissons ratio (n) vs. the temperature (T).

550

J.W. Ringsberg, T. Lindba ck / International Journal of Fatigue 25 (2003) 547558

Fig. 5. The heat capacity (c) and the thermal conductivity (l) vs. the temperature (T).

which alternately operated on the rail head or rail foot to simulate up and down rail bending; see Lindba ck [6] for details and the type of boundary conditions used. An elastic beam model of the rail was used to calculate the displacements caused by the rollers, after which the curvature and contact loads could be estimated. The contact loads were applied to the rail head/rail foot as a distribution of time-dependent nodal forces, see Fig. 2 for the FE mesh. These forces were dened as active, using ramp functions for increasing/decreasing magnitude, while the rail was in contact with a roller; otherwise they were set to zero. The nodal forces had a Hertzian distribution that was calculated according to the Hertz theory for elastic rolling of a cylinder on an innite plane. Note that the Hertz theory presumes elastic deformation in the contact zone. However, the plastic deformation during rail straightening was known to be small.

The yield stress, sy, was dened by a von Mises yield surface, where the yield function, f, was dened as f(s,a)

3. The FE tool developed for RCF analysis of rails An FE tool was developed for the analysis of RCF of railway rails, see Ringsberg [1] for details. It mimics wheelrail rollingsliding contact on any track, since it incorporates both the dynamic global track response and the three-dimensional local elasticplastic contact conditions in the rail head. The FE tool was improved here to include different types of initial stress-state conditions in the rail. As a result, it was possible to apply it in analysing the inuence of residual stresses, caused by the rail manufacturing process, on the fatigue life to crack initiation. Section 3.1 gives a short description of the new FE tool; Section 3.2 species the contact loads of the Iron-ore Line train trafc situation analysed; Section 3.3 presents three analyses with different initial stress-states of the rail. The FE tool analyses were made with the FE code ABAQUS [12]. 3.1. Descriptions of FE tool, and models of track and rail Two FE models, for track and rail, which are coupled by time-dependent boundary conditions form the FE tool. An elastic FE analysis using the track model calculates time-dependent displacements at two cross-sections 12 cm apart; these are then used as boundary conditions in an elastic-plastic FE analysis using the rail model. As a result, the inuences of both the dynamic global track response and the three-dimensional local elastic-plastic material response in the rail are incorporated in the rail fatigue analysis. It was shown in Ringsberg [1] that the global track response predicted by the track model affects the longitudinal residual stress-state in the rail head by some 10 per cent, as compared with a situation where it is disregarded.

|tdev| tdev:tdev In Eq. (1), sdev is the deviator stress tensor of the applied stress tensor, s; a is the backstress tensor; and the operator : denes the contraction a:b = aijbij. The material is elastic when f(s,a) 0 and plastic when f(s,a) = 0. The backstress was determined by da Hdep (2) where H is the hardening modulus and dep is the change in plastic strains. The normality for the plastic ow was dened as df (3) dep l ds is the plastic multiplier. The consistency conwhere l dition was then dened as df df df df :dsH : dl 0 ds ds ds (4)

2 |t
3

dev

|sy with tdev sdeva,

(1)

2.2. Simulation of roller straightening The FE simulation of the roller straightening process required that the curvature of the rail be modelled as prescribed out-of-plane degrees of freedom (DOF) which were time-dependent [11]. These DOF were nally released and the rail was allowed to take its unloaded equilibrium position. In addition to these DOF, the rail was locally loaded on the rail head/rail foot by the rollers in the roller straightening machine. The roller contacts were represented by distributed contact loads,

J.W. Ringsberg, T. Lindba ck / International Journal of Fatigue 25 (2003) 547558

551

The track model is an elastic beam element one which is used in dynamic analysis: 32 sleeper bays are designed to represent an arbitrary track. The ballast material, the pads, the sleepers, and the rail beam are all modelled by nite elements to enable realistic simulations of train trafc conditions. Distributed and moving force combinations of normal, longitudinal, and transverse/lateral rail force components represent a train vehicle travelling along the track. The rail model is a three-dimensional (3D) solid element model. The material volume near the wheel-rail contact undergoes elastic-plastic material response. Hence, high mesh density is used to resolve all the stress and strain gradients satisfactorily. The elasticplastic material behaviour was modelled with the nonlinear kinematic hardening model proposed by Jiang [13,14]; Ekh [15] has coded this model as a user-supplied subroutine to the FE code ABAQUS. In addition, apart from the volume near the wheel-rail contact, the material response is elastic and the mesh is modelled coarsely. Fig. 6 shows the rail model used in the present investigation. Since train trafc on a straight track was investigated, the wheel-rail contact position was at the symmetry line in the rail longitudinal direction; this was veried by inspections at track. The wheel-rail contact loads acting on the rail model, in the normal, longitudinal, and transverse/lateral directions, were represented by moving distributed normal pressure and shear stress distributions. The distributions were calculated according to the Hertz theory of rolling contact between two elastic non-conforming solids with smooth and continuous con-

tact surfaces [16]. The normal load was represented by a Hertzian contact pressure distribution, p(x,y), where (x,y) are the local coordinates that dene the contact zone in the longitudinal and transverse/lateral directions, respectively. The traction forces, q(x,y), were modelled as proportional to p(x,y), that is q(x,y) = m(mx,my)p(x,y) where (mx,my) are the friction coefcients in the (x,y) directions. Hence, regions of stick and slip were disregarded, although they are known to occur. 3.2. Contact loads and rail steel material The Iron-ore Line between Kiruna (Sweden) and Narvik (Norway) is two-way-trafcked on one track, i.e. loaded iron-ore cars run from Kiruna to Narvik where the ore is unloaded before the cars return (unloaded) to Kiruna on the same track. The loaded iron-ore cars have an axle load of 30 tonnes and they run at 50 km/h on a straight track. Since these loaded iron-ore cars contribute the most to RCF damage of the rail, their contact loads were used here. The 30 tonne axle load corresponded to a rail normal force of 147103 N which, according to the Hertz theory for the present wheel and rail geometry, equals a peak normal pressure of 1401 MPa. As straight track was modelled, only the in-plane rail longitudinal force component due to traction was included. The friction coefcient was taken as mx = 0.35, which is a representative value determined by material testing for dry contact for the current wheel and rail materials [4]. Hence, there was a shear stress distribution for fully slipping contact only in the longitudinal direction of the rail. The material for the rail FE model was pearlitic steel grade 1100. Its constitutive material behaviour was represented by the nonlinear kinematic hardening model proposed by Jiang [13,14]. The Jiang model has the capacity to simulate cyclic plasticity and ratchetting material response with decaying rate. This type of constitutive model should be employed for the current repeated wheelrail contact analyses, since a linear kinematic hardening model, for example, cannot describe the material behaviour accurately for the current loading situation. The mechanical properties for the current steel grade are presented in Table 1, while the parameters in the Jiang model used to describe the cyclic behaviour of the material were taken from Ekh [15].

Table 1 Mechanical properties for the pearlitic steel grade 1100 E (GPa) sy (MPa) b c e f (%) 10.3 s f (MPa) 936 ec J

Fig. 6.

The rail model of the FE tool.

209.82 400.1

0.089 0.559

11.5

0.2

552

J.W. Ringsberg, T. Lindba ck / International Journal of Fatigue 25 (2003) 547558

3.3. Application of the FE tool to three examples Three initial stress-state conditions of a rail were analysed with the FE tool: (a) stress-free rail, (b) measured residual stress eld of newly manufactured rails, and (c) calculated residual stress eld from a rail manufacturing simulation. The residual stress eld (b) was measured to validate the computational model used to simulate the rail manufacturing process (c), see Section 2. A user-supplied subroutine was formulated in the FE code ABAQUS, to transfer the results from the 2D FE analysis (Section 2) to the 3D rail FE model. This was done by interpolation, between nodes, of node average stresses in the longitudinal direction. These stresses became initial conditions to the rst step of the 3D rail FE model calculation. No contact loads were applied for this step, and equilibrium was reached after some iterations. Note that material properties for steel grade 900A was used in the rail manufacturing simulation, while material properties for steel grade 1100 was used in the FE tool. The reason was that not all data for the 1100 material was known or found in the literature to make the rail manufacturing simulations for this material. However, these were available for the 900A material. Several rail manufacturing simulations were performed using known data for the steel grade 1100 together with additional steel grade 900 data, to check how the results were affected. It was observed and concluded that the discrepancy in materials does not affect the conclusions drawn in the present investigation, since the residual stresses caused by the rail manufacturing process, for a steel of grade 1100, are of the same magnitude as for 900A and have a similar distribution.

4. Life prediction of RCF crack initiation All materials contain microcracks or defects, etc., that grow during fatigue loading. The fatigue growth of cracks can be studied using fracture mechanics based approaches which are divided into three groups according to length scales [17]: microstructural fracture mechanics (MFM), elastic-plastic fracture mechanics (EPFM), and linear-elastic fracture mechanics (LEFM). The objective of the present investigation, however, emphasises calculating the time to fatigue crack initiation. This could be done using a strain-based total-life approach, for which the growth of fatigue cracks does not need to be modelled. In this approach, it is important to dene what is meant by crack initiation. From an engineering point of view, fatigue crack initiation is often used as the denition of a crack that has completed its Stage I growth, i.e. crack sizes of approx. 5 to 10 grains [17]. The material used here was a pearlitic steel grade 1100. The size of a fatigue-initiated crack in this material, fol-

lowing the above denition, was 0.10.5 mm. Therefore, fatigue crack initiation was dened here as a crack that has a grown to a size of approx. 0.5 mm. Ringsberg [24] proposed a method for fatigue analysis and life prediction of fatigue crack initiation in rolling contacts; Fig. 7 shows the ow chart of the steps in the method. The method was designed as an iterative procedure by which each step should be critically revised until good agreement with test results was achieved. It has been successfully used in previous work to determine the time to fatigue crack initiation in rails and twin disc specimens used for accelerated RCF testing [2]. The results from the numerical analyses, presented in Sections 2 and 3, were analysed for RCF crack initiation using this method; they are presented in Section 5. In the method proposed, it was presumed that the rail material was homogeneous and free from microcracks or defects, etc. The wheelrail rollingsliding contact loads were assumed to cause fatigue damage to the rail due to the elastic-plastic material response in and near the wheel-rail contact zone. Hence, the RCF damage was induced by low-cycle fatigue and ratchetting damage mechanisms. A strain-based total-life approach was adopted in the fatigue life predictions. The strain-life approach used together with FE analyses forms a powerful combination. For example, arbitrary geometries can be analysed for fatigue, for any material and loading, without presuming at the outset the positions and sizes of primary fatigue cracks. Macroscopic material parameters that are determined from (standard) tests can be utilised in the fatigue analysis; they can also describe the material by a constitutive material model in the FE analysis. Hence, a variety of materials can easily be compared in parametric studies by a substitution of material parameters [4]. A fatigue analysis was carried out as a post-process using the results from an FE analysis. Hence, the FE and fatigue analyses were uncoupled. A damage summation rule similar to Palmgren-Miners, presented in Ringsberg [2], was used to calculate fatigue damage in the rail for every wheel passage, n, from low-cycle fatigue (LCF) and ratchetting failure (RF). The number of wheel passages to crack initiation for the current load and wheel for LCF and NRF for ratchetpassage is denoted as NLCF f f ting damage, respectively. A fatigue damage parameter proposed by Jiang [18], see Eq. (5), was used to calculate the fatigue damage caused by low-cycle fatigue for every wheel passage n, for wheel passage n. The material i.e. DLCF = 1 / NLCF f plane with the largest value of the fatigue parameter in Eq. (5a), FPmax, identied the crack plane and Eq. (5b) for the current wheel passage was used to calculate NLCF f n, crack plane, and FPmax as FPmax smax

e Jtg 2

(5a)

max

J.W. Ringsberg, T. Lindba ck / International Journal of Fatigue 25 (2003) 547558

553

Fig. 7.

A method for fatigue life prediction of crack initiation [4].

FPmax

2 (s f) LCF b+c )2b s ) (2NLCF f e f (2Nf f E

(5b)

In Eq. (5), denotes the MacCauley bracket, x = 0.5( |x| + x); smax and e are the maximum stress and the strain range normal to the crack plane; t and g are the shear stress range and the (engineering) shear strain range on the crack plane; J is a material and load dependent constant; E is the elastic modulus; s f and e f are the axial fatigue strength and the axial fatigue ductility coefcients; b and c are the fatigue strength and the fatigue ductility exponents. The energy-density based model in Eq. (5) was used together with the critical plane concept to identify the position and orientation of the crack plane. Damage caused by ratchetting was computed for every wheel passage, n, as DRF = 1 / NRF f . A criterion for ratchetting failure as proposed by Kapoor [19] was used as for this purpose to calculate NRF f ec / er where er NRF f

Eqs. (5) and (6), for the pearlitic steel grade 1100 which has properties to the pearlitic steel grade BS11 used in previous work. Note, however, that the yield stress for the pearlitic steel grade 1100 is normally 690 MPa. The lower value of the yield stress in Table 1 was the yield stress used in the current constitutive material model and it was obtained by parameter optimisation to t experimental data for ratchetting material behaviour from cycle 1 to 600, see Ekh [15] for details.

5. Results 5.1. Rail manufacturing simulation The contour plots of the calculated residual stress eld in the longitudinal direction, sx, after cooling and roller straightening can be seen in Figs. 8 and 9, respectively.

)2 g / 3 (e

(6)

In Eq. (6), ec is a material constant that represents the material ductility when it is subjected to rolling contact loading [2]; er is an equivalent ratchetting strain per , wheel passage estimated from the average normal, e , ratchetting strains per wheel passage. The and shear, g ratchetting deformation in rolling contacts is, in most instances, governed by the ratchetting shear strain. Damage summation was calculated for every wheel passage, n, as D

n1

Nf

max((dDLCF / dN)n, (dDRF / dN)n)

(7)

The proposed damage summation rule accounts for a varying damage rate in the damage summation, which may be caused by, for example, a decaying ratchetting rate due to constant or variable amplitude loads. Failure is assumed to occur when D = 1. Table 1 presents the mechanical properties used in

Fig. 8. Contour plot of the rail longitudinal residual stress distribution (sx) after cooling.

554

J.W. Ringsberg, T. Lindba ck / International Journal of Fatigue 25 (2003) 547558

agreement in the rail foot was related to the rather simple material model (linear kinematic hardening model) used in the rail manufacturing simulation. This material model could not describe the material behaviour properly, during cyclic loading, in the roller straightening simulation, see Schleinzer [20]. A nonlinear kinematic hardening model was recommended for use in future work to achieve better agreement between measurements and simulation in the entire rail cross-section. However, as RCF analysis of the rail head was the main interest of the present investigation, the poor agreement in results for the rail foot had no inuence on the results or conclusions from the subsequent FE tool and RCF analyses. 5.2. FE tool and RCF analysescomparison of initial conditions to rail FE model Eight wheel passages were simulated using the FE tool. The material response in the rail head was ratchetting, which decayed for every wheel passage. The RCF assessments were made following the proposed evaluation method given in Section 4. The three FE tool analyses were compared with respect to the position of the greatest fatigue damage, the orientation of the material plane most damaged (crack plane), and the fatigue life to crack initiation on this plane. Table 2 presents results from the three FE tool and RCF analyses. The results are presented for the eighth wheel passage alone: the maximum residual von Mises effective stress (smax eff ); the maximum accumulated effective plastic strain according to von Mises (emax eff, pl); the maximum value of the fatigue parameter (FPmax), recorded during the eighth wheel passage on a crack plane dened by an orientation j and by the distance d from the rail head surface (see Fig. 11). This type of comparison revealed whether the initial stress eld of the rail model was redistributed, or if fatigue damage accumulation was still inuenced by the initial conditions of the model. In all of the three FE tool analyses d was calculated as 2.8 mm, i.e. the greatest damage was subsurface. This was expected due to the very high axle load and the low and friction coefcient. The largest values of smax eff were also found at this position. The orientation emax eff, pl of j was in accordance with that observed in damaged rails that had been taken out from service for inspection. Fig. 12 shows the calculated residual von Mises effective stress in the rail head after the rail manufacturing simulation, i.e. it shows the initial residual stress distribution to example (c). The residual von Mises effective stress after the eighth wheel passage is presented in Fig. 13. A contour plot of the accumulated effective plastic strain according to von Mises after the eighth wheel passage is shown in Fig. 14, and Fig. 15 shows the accumulation of this strain for every wheel passage for the examples (a) and (c).

Fig. 9. Contour plot of the rail longitudinal residual stress distribution (sx) after cooling and roller straightening.

Note that the residual stress-state calculated in the rail after cooling (Fig. 8) seems to have no effect on the residual stress-state (Fig. 9) after roller straightening. The result in Fig. 9 is also shown in Fig. 10 as a graph that shows the calculated stress distribution together with results from residual stress measurements made in newly manufactured rails. The discontinuities in the graph, especially in the web region, are associated with numerical errors from the calculation. The measurements were made on the surface of a rail using a hole-drilling strain gauge method [11]; the measurement uncertainty according to this method was 10 per cent. The agreement was good in the rail head but poor in the rail foot. The dis-

Fig. 10. Calculated (solid line) and measured residual stresses in newly manufactured rails.

J.W. Ringsberg, T. Lindba ck / International Journal of Fatigue 25 (2003) 547558

555

Table 2 Results from three FE tool and RCF analyses for the eighth wheel passage Examples (a) Stress-free; reference example (b) Measured stress distribution (c) Calculated stress distribution smax eff (MPa) 270 265 273 emax eff, pl (%) 1.62 1.63 1.69 FPmax (MPa) 1.32 1.25 1.33 j (deg.) 40 41 40 d (mm) 2.8 2.8 2.8

Fig. 14. The accumulated effective plastic strain according to von Mises after the eighth wheel passage. Fig. 11. Denition of the crack plane angle j and the distance d to the position of the greatest fatigue damage.

Fig. 12. The residual von Mises effective stress in the rail head after the rail manufacturing simulation. Fig. 15. The maximum value of the accumulated effective plastic strain according to von Mises vs. the number of wheel passages.

Fig. 13. The residual von Mises effective stress in the rail head after the eighth wheel passage.

The results in Table 2 show that the difference between the three examples was small, which indicates that the residual stresses introduced initially were already redistributed during the rst wheel passages. In relation to the reference example, the largest percentage difference in the maximum von Mises effective stress was 2%; the maximum effective plastic strain according to von Mises, 4%; the maximum fatigue parameter recorded during the eighth wheel passage, 5%; there was a 2% difference in j. A calculation of the train trafc time (in days) to subsurface fatigue crack initiation was made using the results from the complete RCF analyses and statistics for

556

J.W. Ringsberg, T. Lindba ck / International Journal of Fatigue 25 (2003) 547558

the annual train trafc density at the Iron-ore Line. Note here that the residual stress in the rail head for the examples (b) and (c) in Fig. 10 is of almost the same magnitude. The small discrepancy gives rise to a difference in the fatigue parameter FPmax during wheelrail contact loading, and hence, the time to fatigue crack initiation is also inuenced. In addition, the damage rate for subsequent wheel passages was extrapolated using the values calculated for the sixth to eighth wheel passages. As a result, subsurface fatigue crack initiation was determined to be 2 days of train trafc for the reference example (a), example (b) showed 19% more time to fatigue crack initiation, and example (c) 2% less. These results have not been validated because the fatigue cracks were subsurface-initiated, which meant it was not possible to acquire results for validation from visual inspections.

Fig. 16. Longitudinal residual stress (sx) after eight wheel passages vs. axle load.

6. Discussion The inuence of residual stresses caused by the rail manufacturing process on the rail fatigue life to crack initiation was investigated. A comparison of three examples showed that the initial residual stress-state does not have any great inuence on the fatigue crack initiation life for the train trafc situation analysed. It was concluded that the local wheel-rail contact load governs fatigue damage, in particular for very high axle loads. Note, however, that the RCF life prediction method identied the greatest fatigue damage at the depth where the start of fatigue cracking was found in cracked rails taken out of service. The discrepancy in magnitude of the longitudinal stresses in the rail head, between examples (b) and (c) shown in Fig. 10, was within the error margin of the measurements. Note that this error was transferred throughout the numerical analyses, see the results in Table 2. The simulation of the rail manufacturing process can be improved in future work. In the present investigation, the roller straightening process was simulated using a 2D FE model. Schleinzer [5] simulated the roller straightening process in three-dimensional FE calculations. An advantage of using Schleinzers 3D FE model is that all stress and strain components are incorporated in the FE calculation. On the other hand, the computational time is excessive when compared with the 2D FE model used in the present investigation. Using the Schleinzer approach would not change the conclusions drawn in the present investigation, since the local wheelrail contact load still governs the rail fatigue damage. The fatigue damage accumulated in the rail from the roller straightening process was not incorporated in the present investigation. Here, only the stress-state caused by the cooling and roller straightening steps was transferred to the rail model of the FE tool where it was used as an initial condition. The reason it was disregarded

5.3. When do initial residual stresses affect the rail fatigue life?

The results presented for the 30 tonne axle load in Section 5.2 showed that the initial residual stress-state had negligible inuence on the fatigue life to crack initiation. The question is then if, or when, an initial residual stress-state, here caused by the rail manufacturing process, affects the fatigue initiation life. According to the results presented in Fig. 10, the longitudinal residual stress in the rail head after rail manufacturing is tensile. The initiation of surface-initiated cracks called head checks is governed by a ratchetting shear strain on the rail head surface, see Ringsberg [4]. However, the fatigue initiation life is shorter when a tensile stress acts normal to the crack plane, see Eq. (5), than it is when there is compressive stress. Consequently, the fatigue life to crack initiation of head checks can be expected to be lower, for example (c), as compared with (a). The FE tool was utilised for six axle loads to determine the axle load for which the longitudinal residual stress changes from tension to compression. Examples (a) and (c) were compared in separate FE tool calculations for the axle loads 10, 12, 15, 20, 25, and 30 tonnes, with mx = 0.35. The calculations were done for eight wheel passages, and the results are presented in Fig. 16 as the longitudinal residual stress after the eighth wheel passage. The results show that the longitudinal residual stress changes from tension to compression at a 13 tonne axle load, for example (a), and at 16 tonnes, for example (c). For both of these axle loads, the greatest fatigue damage was found at the rail head surface for the prevailing contact load conditions. The conclusion was that, for an axle load of the same magnitude, example (c) has a shorter fatigue initiation life than example (a).

J.W. Ringsberg, T. Lindba ck / International Journal of Fatigue 25 (2003) 547558

557

was that, in the authors opinion, its contribution to total accumulated fatigue damage is negligible. However, for complete RCF analysis of rail fatigue life, fatigue damage accumulation calculations could start by recording the accumulation of plastic strain from the rail manufacturing process. To keep track of changes in residual stress, plastic strain, material parameters, and the evolution of fatigue damage, without loss of information between the different steps, would require using the same 3D rail FE model and the same constitutive material model through the entire simulation. The present work emphasised on RCF evaluation of rails at straight track, where the contact load motion was along the symmetry line in the longitudinal direction. At curved track, the contact position/positions are closer to the rail head gauge corner. The tangential loads and normal contact load are often higher, as compared with those at straight track, for the same train trafc conditions. As a result, the initial residual stresses caused by the rail manufacturing will almost immediately redistribute, and thus, have less inuence on the RCF life to crack initiation. The results presented in Fig. 16 illustrate how the axle load/normal contact load affects the longitudinal residual stress. In addition, the magnitude of the friction coefcient used to model the tangential loads in the longitudinal direction is important for the outcome of the results. A higher value would have increased the plastic shear deformation near the surface and reduced the time to fatigue initiation, while a lower value reduces surface plastic ow and the risk/time for surface fatigue of rails.

compared with residual stresses measured in newly manufactured rails. The FE tool and fatigue analyses of the three examples showed that there was little difference in results with respect to position of the greatest fatigue damage, orientation of the crack plane, and time to fatigue crack initiation. The initially introduced stress-states of the rail model in the FE tool were already redistributed after the rst wheel passages. Hence, it was concluded that the local wheel load governs the fatigue damage of rails. The position of the greatest fatigue damage was calculated to be 2.8 mm below the surface. This depth corresponded to the depth where cracks have been found to initiate. The FE tool calculations made for several axle loads, for example (a) and (c), showed that the longitudinal residual stress caused by the rail manufacturing process shortens the onset of fatigue initiation of surfaceinitiated cracks, i.e. head checks. In future work, the fatigue damage calculations could also incorporate the damage introduced during the rail manufacturing process. Although this contribution may be negligible in total, it offers a more complete RCF life prediction model which can be important in assessments of different types of rail materials and train trafc situations. Calculating this damage would require using the same 3D rail FE model and the same constitutive material model through the entire simulation, to incorporate the multiaxial effects of the evolution of fatigue damage.

7. Conclusions Acknowledgements The cooling and roller straightening sequences of the rail manufacturing process were simulated in this investigation. The objective was to calculate the residual stress eld caused by the rail manufacturing process and to determine whether this inuences the RCF life of rails. Three examples of initial conditions in a rail were compared using an FE tool developed for RCF analysis of rails: (a) stress-free rail, (b) introduction of a measured residual stress-state in a newly manufactured rail, and (c) introduction of a residual stress-state as calculated from a rail manufacturing simulation. With the FE tool and fatigue calculations, the three examples were compared. The three sets of initial conditions were designed in accordance with the heavy-haul train trafc situation at the Iron-ore Line (in Sweden) where subsurface crack initiation has been observed. The conclusions can be summarised as follows. The results from the numerical simulation of a rail manufacturing process showed acceptable agreement for the longitudinal residual stress eld in the rail head, which was the main interest in this investigation, when The work presented was funded by the Swedish National Board for Industrial and Technical Develop Railway ment (NUTEK, now VINNOVA), Lulea Research Centre (JVTC), the Swedish National Centre of Excellence CHARMEC (CHAlmers Railway UniMECHanics), and the Polhem Laboratory at Lulea versity of Technology. Professor Lennart Josefson at the Department of Applied Mechanics, Chalmers University of Technology, and Mats Na sstro m at the Division of University of TechComputer Aided Design, Lulea nology, are acknowledged.

References
[1] Ringsberg JW, Bjarnehed H, Johansson A, Josefson BL. Rolling contact fatigue of railway railsnite element modelling of residual stresses, strains and crack initiation. Proc IMechE, Part F, J Rail Rapid Transit 2000;214:719.

558

J.W. Ringsberg, T. Lindba ck / International Journal of Fatigue 25 (2003) 547558

[2] Ringsberg JW, Josefson BL. Finite element analyses of rolling contact fatigue crack initiation in railheads. Proc IMechE, Part F, J Rail Rapid Transit 2001;215:24359. [3] Ringsberg JW. Cyclic ratchetting and failure of a pearlitic rail steel. Fatigue Fract Engng Mater Struct 2000;23:74758. [4] Ringsberg JW. Life prediction of rolling contact fatigue crack initiation. Int J Fatigue 2001;23:57586. [5] Schleinzer G, Fischer FD. Residual stress formation during the roller straightening of railway rails. Int J Mech Sci 2001;43:228195. ck T, Na sstro m MO. Residual stresses in railway rails after [6] Lindba manufacturing. In: Mori K, editor. Proceedings 7th International Conference on Numerical Methods in Industrial Forming Processes (NUMIFORM). Lisse: Balkema; 2001. p. 5736. [7] Webster PJ, Wang X, Mills G, Webster GA. Residual-stress changes in railway rails. Physica B 1992;180:102931. [8] Urashima C, Sugino K, Nishida S. Generation mechanism of residual stresses in rails. In: Proceedings 3rd International Conference on Residual Stresses (ICRS-3). Oxford: Elsevier Science; 1991. p. 148994. sstro m MO, Webster PJ, Wang J. Residual stresses and defor[9] Na mations due to longitudinal welding of pipes. In: David SA, Vitek JM, editors. Proceedings 3rd International Conference on Trends in Welding Research. Materials Park, OH: ASM International; 1992. p. 10915. [10] Kreith F, Bohn MS. Principles of heat transfer. Pacic Grove, CA: Brooks/Cole, 2001.

ck T. Computational modelling of rail manufacturing. [11] Lindba Licentiate of Engng Thesis, Division of Computer Aided Design, University of Technology, Lulea , Sweden, 2000. Lulea [12] Hibbitt, Karlsson & Sorensen. ABAQUS/Theory Manual, version 5.8. Pawtucket, RI: Hibbitt, Karlsson & Sorensen, 1998. [13] Jiang Y, Sehitoglu H. Modeling of cyclic ratchetting plasticity, part I: development of constitutive relations. Trans ASME J Appl Mech 1996;63:7205. [14] Jiang Y, Sehitoglu H. Modeling of cyclic ratchetting plasticity, part II: comparison of model simulations with experiments. Trans ASME J Appl Mech 1996;63:72633. [15] Ekh M, Johansson A, Thorberntsson H, Josefson BL. Models for cyclic ratchetting plasticityintegration and calibration. Trans ASME J Engng Mater Technol 2000;122:4955. [16] Johnson KL. Contact mechanics. Cambridge: Cambridge University Press; 1985. [17] Miller KJ. The three thresholds for fatigue crack propagation. In: Piascik RS, Newman JC, Dowling NE, editors. Fatigue and Fracture Mechanics: 27th volume ASTM STP 1296. Philadelphia, PA: American Society for Testing and Materials; 1997. p. 26786. [18] Jiang Y, Sehitoglu H. Rolling contact stress analysis with the application of a new plasticity model. Wear 1999;191:3544. [19] Kapoor A. A re-evaluation of the life to rupture of ductile metals by cyclic plastic strain. Fatigue Fract Engng Mater Struct 1994;17:20119. [20] Schleinzer G, Fischer FD. Residual stresses in new rails. Mater Sci Engng 2000;A288:2803.

Vous aimerez peut-être aussi