Vous êtes sur la page 1sur 11

Wear 253 (2002) 498508

A comparison of analytical and numerical methods for the


calculation of temperatures in wheel/rail contact

Martin Ertz

, Klaus Knothe
Technische Universitt Berlin, Sekr. F5, Marchstr. 12, D-10587 Berlin, Germany
Received 3 January 2002; received in revised form 11 April 2002; accepted 1 May 2002
Abstract
The maximum surface temperature during rolling contact of railway wheels with sliding friction can be estimated using Bloks ash
temperature formula. For a more detailed investigation, semi-analytical and numerical methods are available. A survey of various methods
is given and an efcient approach is proposed for Hertzian contact. The actual contact temperature is conned to a very thin surface layer.
Due to continuous frictional heating, the bulk temperature of the wheel increases with time. For the long-term behaviour of the wheel
temperature, not only the convection at the free wheel surfaces but also the heat conduction from the wheel into the colder rail has to be
considered. Practical consequences of the theoretical results are discussed.
2002 Elsevier Science B.V. All rights reserved.
Keywords: Wheel/rail contact; Flash temperature; Contact temperature; Convection
1. Introduction
The slip between wheel and rail causes frictional heating
of both bodies. The resulting high contact temperatures have
to be taken into account for microstructural alterations, such
as formation of white-etching layers (WEL) on rail surfaces
and also for failure of wheel and rail. With a possible relation
between contact temperature and coefcient of friction, they
can also be responsible for decreasing creep force curves.
Since contact temperatures under real railway operating con-
ditions can hardly be measured, temperature calculation is
an important part in the investigation of rolling contact.
Fundamental work in this area has been done by Blok [1]
and Jaeger [2]. In his article The ash temperature concept
published in 1963, Blok gave a survey of the state-of-the-art
at that time [3]. The methods used are based on the theory
of heat conduction with moving heat sources as described
by Carslaw and Jaeger [4]. Archard [5] used the theory of
Blok and Jaeger to investigate some tribological problems.
An approximate solution for rolling contact was given by
Tanvir [6]. Knothe and Liebelt [7] presented a numerical
method for arbitrarily distributed heat sources.

Dedicated to the memory of Prof. Harmen Blok (19102000), the


pioneer of ash temperature calculation.

Corresponding author. Tel.: +49-30-314-22142;


fax: +49-30-314-22866.
E-mail address: martin.ertz@tu-berlin.de (M. Ertz).
While railway wheels are heated by friction in the con-
tact patch, there is also heat loss due to conduction through
the contact patch into the rail. This effect has rarely been
taken into account in previous research. It was mentioned
by Moyar and Stone [8], and Gupta et al. [9] used a sim-
ple approach for the approximate consideration. In addition,
heat ows from the wheel into ambient air due to convec-
tion at the free surfaces. Cameron et al. [10] calculated a
steady-state temperature based on the equilibrium of fric-
tional heating and heat loss due to convection. Fischer et al.
[11] found an analytical solution for the surface temperature
with combined frictional heating and convection.
All these works are conned to smooth surfaces and most
of these are based on the theory of Hertz for the mechan-
ical contact problem. But real surfaces are always rough.
Using measured proles of rough surfaces, even the calcula-
tion of stress and strain demands computer simulations [12].
With smooth surfaces, the heat conduction can be treated
one-dimensionally, but for rough surfaces, this simplica-
tion is not possible, and the investigation of temperatures
is also a very complicated numerical problem [13]. On the
other hand, Archard [5] has already stated that the inuence
of surface roughness can be neglected to a rst approxima-
tion, since the largest temperatures are those deduced for
the whole region rather than those deduced for the smaller
individual contact areas.
In the following sections, methods are presented for tem-
perature calculation with smooth surfaces. After a short
0043-1648/02/$ see front matter 2002 Elsevier Science B.V. All rights reserved.
PII: S0043- 1648( 02) 00120- 5
M. Ertz, K. Knothe / Wear 253 (2002) 498508 499
introduction to the theory of fast moving heat sources, the
temperature distribution in wheel/rail contact will be investi-
gated in detail. Convection and heat conduction through the
contact patch are also taken into account for the long-term
behaviour of the wheel temperature. The possible inu-
ence on thermally induced phase transformations will be
discussed.
2. Moving heat sources in rolling contact
When wheel and rail are brought into contact under the
action of the static wheel load, the area of contact and the
pressure distribution are usually calculated with the Hertzs
theory. In this case, the area of contact is elliptical and the
normal pressure distribution is [14]
p
z
(x, y) = p
0
_
1
x
2
a
2

y
2
b
2
, (1)
with the maximum pressure
p
0
=
3N
2ab
, (2)
for the normal load N and the semi-axes a (in rolling direc-
tion) and b of the contact ellipse (Fig. 1). In the case of a
innitely long cylinder subjected to the normal load N/b

per unit length, the normal pressure distribution is [14]


p
z
(x) = p
0
_
1
x
2
a
2
, (3)
with the maximum pressure
p
0
=
2N
ab

. (4)
This model is often used for a simplied analysis of
three-dimensional contact problems. Comparing Eqs. (2)
and (4), we obtain the reference length b

for the transition


from the three-dimensional to the two-dimensional case as
b

=
4
3
b. (5)
Fig. 1. Elliptical area of contact.
Fig. 2. Co-ordinate system for temperature calculation in wheel/rail con-
tact.
If a tangential force T is transmitted between wheel and
rail, there is always a mean relative velocity in the contact
point. High contact temperatures are to be expected only
with the transmission of tractive or braking forces at high
relative velocities. In this case, sliding occurs within the
whole contact area and the tangential force is T = N.
The coefcient of friction is assumed to be constant in
the following.
From the point of view of an observer who is xed to the
wheel, the contact patch moves with respect to the wheel
surface and the frictional heating within the contact patch is
a time-dependent heat source (Fig. 2). During the very short
time period that every point on the surface is in contact, the
thermal penetration depth
=
a

L
, (6)
is very small compared to the size of the contact patch. It
depends on the non-dimensional Pclet number
L =
av
2
, (7)
with the semi-axis length a, the speed v of the moving heat
source and the thermal diffusivity
=

c
, (8)
that combines the material properties (thermal conductiv-
ity), (density) and c (specic heat capacity). Since the ve-
locities of wheel and rail with respect to the contact patch
are different, this also has to be taken into account for the
Pclet number and the thermal penetration depth. As stated
by Johnson, L may be interpreted as the ratio of the surface
speed to the rate of diffusion of heat into the solid [14]. If
L > 10, heat conduction occurs only perpendicular to the
contact plane, i.e. in z-direction [2,5,7]. With the typical val-
ues for wheel/rail contact, a 5 mm, = 14.210
6
m
2
/s
[9] and v
0
= 30 m/s, one gets L = 5300. The longitudinal
and lateral heat conduction (x- and y-direction) can, there-
fore, be neglected and the heat conduction equation is [15]

z
2
=

t
. (9)
500 M. Ertz, K. Knothe / Wear 253 (2002) 498508
Since Eq. (9) is linear and contains only derivatives of the
temperature, the zero-point of temperature may be conve-
niently chosen as the ambient temperature. Thus, repre-
sents the temperature rise due to the heat supply within the
contact patch.
We rst consider the case that wheel and rail are initially
at ambient temperature on coming into contact. Inside the
contact patch they are subjected to a heat source at their
surfaces due to the frictional heating. Since the heat ow is
one-dimensional, this problem is similar to a semi-innite
solid with an arbitrarily distributed heat source q(t ) applied
to the surface z = 0 at t 0. The solution (z, t ) has to
full the differential Eq. (9), the initial condition
(z, t = 0) = 0, (10)
and the boundary condition

z
(z = 0, t ) = q(t ). (11)
The general solution of this problem can be found in the
book of Carslaw and Jaeger [4], Section 2.9,
(z, t ) =
1

_
t
0
q(t t

) exp
_

z
2
4t

_
dt

, (12)
with the thermal penetration coefcient
=
_
c =

. (13)
This is usually referred to as b in the literature [15]. To
avoid confusion with the length of the lateral semi-axis of
the contact ellipse, we use instead.
The heat conduction Eq. (9) can also be solved by means
of the Laplace transform for special boundary conditions
(e.g. Tanvir [6], Knothe and Liebelt [7] and Fischer et al.
[11]). But when the distribution of q(t ) is a prescribed
function of time, the temperature can easily be found by
solving the denite integral in Eq. (12) using analyti-
cal or numerical methods. Therefore, this method should
generally be preferred since the mathematical effort is
smaller than with the Laplace transform and the results are
equal.
Points on the surfaces of wheel and rail pass through the
contact patch at different speeds due to the sliding velocity.
Since the largest heat ux and the highest temperatures occur
along the major axis which is parallel to the rolling direction
at y = 0, it is most important to examine this case [6]. It
is meaningful to substitute the time t elapsed since entering
the contact patch with the current position x in a co-ordinate
system xed to the contact patch (Fig. 2),
x = vt a. (14)
For a further simplication, we introduce the dimensionless
co-ordinates
=
x
a
and =
z

. (15)
Using Eq. (12) with the respective values of , v and q,
the temperatures of wheel (w) and rail (r) can be calculated
as

w
(, ) =
1

w
_
a
v
w
_

1
q
w
(

)
exp
_


2
2(

)
_
d

, (16)
for the wheel with v
w
= v
0
+v
s
and

r
(, ) =
1

r
_
a
v
r
_

1
q
r
(

)
exp
_


2
2(

)
_
d

, (17)
for the rail with v
r
= v
0
(Fig. 2). The analytical solution
of the integral in Eq. (16) is quite simple if we assume a
constant heat ow rate q
w
at the wheel surface within the
contact patch. For 1 1, we get

w
(, ) =
q
w

w
_
2a
v
w
_
_
2( +1)

exp
_


2
2( +1)
_
erfc
_

2( +1)
_
_
, (18)
with the complement erfc(s) of the error function erf(s) de-
ned as [15]
erfc(s) = 1 erf(s) = 1
2

_
s
0
e

2
d. (19)
Outside the contact area there is no frictional heating. Ne-
glecting convection, the heat ow rate is zero here and the
analytical solution for > 1 is

w
(, ) =
q
w

w
_
2a
v
w

__
_
2( +1)

exp
_


2
2( +1)
_
erfc
_

2( +1)
_
_

_
_
2( 1)

exp
_


2
2(1)
_
erfc
_

2( 1)
_
__
.
(20)
With = 0 in Eqs. (18) and (20), the surface temperature is

w
() =
2 q
w

w
_
a
v
w
_
+1 for 1 1 (21)
and

w
() =
2 q
w

w
_
a
v
w
_
_
+1
_
1
_
for > 1.
(22)
M. Ertz, K. Knothe / Wear 253 (2002) 498508 501
Table 1
Comparison of mean and maximum temperatures for different distributions of frictional heating
Heat ow rate
mean
/(v
s
p
0

a/v
w
/
w
)
max
/(v
s
p
0

a/v
w
/
w
)
max
Elliptic (numerical solution) 0.851 1.235 0.652
Elliptic (improved approximation) 0.851 1.240 (0.4%) 0.643
Elliptic (Tanvirs approximation) 0.853 (0.2%) 1.276 (3.3%) 0.673
Parabolic 0.860 (1.1%) 1.303 (5.5%) 0.5
Constant 0.836 (1.8%) 1.253 (1.5%) 1.0
3. Frictional heating in wheel/rail contact
In principle, the integral in Eq. (16) can be solved an-
alytically for any distribution of heat ow rate given as a
polynomial. For the elliptical heat ow rate in Hertzian con-
tact, this is not the case. The solution for constant heat ow
rate is used as a rst-order estimation in Section 3.1. For a
more detailed investigation, we will introduce a numerical
solution in Section 3.2 and an analytical approximation in
Section 3.3. The results of all calculations are summarized
in Table 1.
3.1. Analytical solution for constant heat ow rate
With constant values of the coefcient of friction and
the sliding velocity v
s
, the frictional power dissipation rate
in the contact patch is proportional to the pressure:
q
friction
() = v
s
p
z
() = v
s
p
0
_
1
2
. (23)
It is generally assumed that all the frictional power dissipa-
tion is transformed in heat. The heat generated in the con-
tact patch ows into the material of wheel and rail. With the
heat partitioning factor , this can be written as
q
w
() = q
friction
()
and q
r
() = (1 ) q
friction
(). (24)
The surface temperatures of wheel and rail must be equal
everywhere in the contact patch. Substituting Eq. (24) in
Eqs. (16) and (17), the part of the frictional heating that
ows into the wheel is
=

w

v
w

v
w
+
r

v
0
. (25)
With the average heat ow rate at the surface of the wheel,
q
w
=
1
2
_
1
1
q
w
() d =

4
v
s
p
0
, (26)
the solutions for constant heat ow rate, Eqs. (18)(22),
can be used as a simple estimate for the friction-induced
temperature of wheel and rail. The maximum temperature

max
= 1.253
v
s
p
0

w
_
a
v
w
, (27)
occurs at the trailing edge of the contact patch (Fig. 3). This
result has already been given by Blok.
3.2. Numerical solution for arbitrary heat ow rate
An arbitrarily distributed heat source can be approximated
by a step-wise constant function. As a modication of the
solution for a constant heat ow rate (Eq. (20)), we get

w
(
k
, ) =
1

w
_
2a
v
w
k

i=1
q
i
{G(
k

i1
, ) G(
k

i
, )}, (28)
with
G(, ) =
_
2

exp
_

2
2
_
erfc
_

2
_
. (29)
Setting = 0 in Eq. (28), the surface temperature is [7]

w
(
k
) =
2

w
_
a
v
w
k

i=1
q
i
_
_

k

i1

k

i
_
.
(30)
Using this method, the surface temperature has been calcu-
lated for various distributions of frictional heating (Fig. 3),
although the case of parabolic heat ow rate could also be
solved analytically. Since the average heat ow rate is equal
in all cases, the maximum temperatures are almost the same
(Table 1). Outside the contact patch, the differences can be
Fig. 3. Surface temperature for various distributions of frictional heat ow
rate.
502 M. Ertz, K. Knothe / Wear 253 (2002) 498508
Fig. 4. Comparison of numerical solution and Tanvirs approximation for
Hertzian contact.
neglected. As shown in Fig. 4, the high temperatures are
conned to a very thin surface layer. With the parameters
given in Table 2, the thermal penetration depth, Eq. (6), is
= 75 m whilst the semi-axis length in rolling direction
is a = 5.88 mm.
3.3. Analytical approximation for Hertzian contact
Tanvir proposed an alternative approach for Hertzian con-
tact [6]. He approximated the heat source distribution for
Hertzian contact (Eq. (23)) to a fourth-order polynomial and
found the analytical solution of Eq. (9) with the Laplace
transform. The temperature outside the contact patch was
approximated to the solution for the constant average heat
ow rate (Eq. (22)). It should be mentioned that some coef-
cients in the Eqs. (15) and (23) of Tanvirs paper are wrong.
They have been corrected for the results shown in Fig. 5. As
Table 2
Reference data for the presented results
Symbol Value Unit Equation
Semi-axis of the contact ellipse
In rolling direction a 5.88 mm
In lateral direction b 10.54 mm
Reference length b

14.05 mm Eq. (5)


Specic heat capacity [9] c 450 J/kg K
Normal force (wheel load) N 100 kN
Maximum Hertzian pressure p
0
770 MPa Eq. (2)
Wheel radius r
0
0.5 m
Vehicle speed v
0
30 m/s
Sliding velocity (longitudinal) v
s
1 m/s
Thermal penetration coefcient 13290 Ws
0.5
/Km
2
Eq. (13)
Thermal diffusivity 14.2 10
6
m
2
/s Eq. (8)
Thermal conductivity [9] 50 W/Km
Coefcient of friction 0.3 1
Density 7850 kg/m
3
These parameters hold for all calculations unless other values are given.
Fig. 5. Temperature distribution in normal direction at the trailing edge
( = 1) and before renewed contact after one revolution of the wheel.
shown before, the results for different distributions of fric-
tional heating are nearly similar outside the contact patch.
Even so, one gets a discontinuity at the trailing edge. The
approximation of the heat ow rate can be improved with a
least square t of the Hertzian pressure distribution. Using
again a fourth-order polynomial, this can be written as
q
friction
() = v
s
p
0
f (), (31)
with
f () =

2048
(645 210
2
315
4
) (32)
for 1 1. The surface temperature for the approx-
imated heat ow rate can be calculated from Eq. (16) for
= 0 as

w
() =
v
s
p
0

w
_
a
v
w
F(), (33)
M. Ertz, K. Knothe / Wear 253 (2002) 498508 503
with
F() =
_

1
f (

)
d

. (34)
This solution can also be used outside the contact area. We
get
F() =

128
_
+1(71 +12 20
2
+8
3
16
4
)
(35)
for 1 1 and
F() =

128
_
_
+1(71 +12 20
2
+8
3
16
4
)

_
1(71 12 20
2
8
3
16
4
)
_
(36)
for > 1. The maximum temperature

max
= 1.240
v
s
p
0

w
_
a
v
w
(37)
is nearly equal to the result of Eq. (27). The local differ-
ences between the improved approximation and the numer-
ical solution for Hertzian contact are everywhere less than
1% (Table 1). Therefore, this efcient approach should be
the best choice for the investigation of contact temperatures
and thermal stresses in the case of Hertzian contact.
4. Heat conduction from wheel into rail
The bulk temperature of the wheel increases with time due
to the continuous frictional heating on its surface. Therefore,
the temperatures of wheel and rail are different when a point
on the surface of the wheel comes into the area of contact
again. This gives rise to a considerable heat ow from the
hot wheel into the cold rail due to conduction through the
contact patch.
After one revolution of the wheel, the temperature gradi-
ent in z-direction is very small (Fig. 5). Therefore, the next
contact of a point on the surface of the wheel with the rail
can be treated as the contact of two semi-innite bodies with
different initial temperatures that come into contact at t = 0
[4,15]. With the initial temperatures of wheel and rail,

w
(z, t = 0) =
w0
and

r
(z, t = 0) =
r0
= 0, (38)
the surface temperature goes instantaneously to a constant
value
m
. Without frictional heating at the same time, the
wheel temperature is (Carslaw and Jaeger [4], Section 2.5)

w
(z, t ) =
m
+(
w0

m
)erf
_
z
2

w
t
_
, (39)
as long as the prescribed constant surface temperature is
m
.
With the dimensionless co-ordinates dened in Eq. (15), the
temperature can be written for the wheel as

w
(, ) =
m
+(
w0

m
)erf
_

2( +1)
_
, (40)
and for the rail as

w
(, ) =
m
_
1 erf
_

2( +1)
__
. (41)
The resulting heat ow rate through the contact patch is
q
w
() =
w
(
w0

m
)
_
v
w
a( +1)
(42)
for the wheel and
q
r
() =
r

m
_
v
0
a( +1)
(43)
for the rail. The value of the constant surface temperature
m
can be calculated from the condition that the heat ow rates
at the surfaces of wheel and rail must be equal everywhere
in the contact patch. From Eqs. (42) and (43) follows:

m
=

w

v
w

v
w
+
r

v
0

w0
=
w0
, (44)
with as dened in Eq. (25). The heat ow rate through the
contact patch from the hot wheel into the cold rail is then
q
w
() =
r

w0
_
v
0
a( +1)
= q
r
(), (45)
and the total heat ow per unit width is

Q
rail
/b

= a
_
1
1
q
w
() d =
r

w0
_
8av
0

. (46)
Outside the contact patch there is no heat ow from the
wheel into the rail. The surface temperature of the wheel is

w
(, = 0)
=
w0
_
1
2

(1 )arcsin
_
2
+1
_
(47)
for > 1 [4]. In this case, an analytical solution for 0
is not possible. It is not strictly necessary to solve the prob-
lem numerically since the analytical solution is sufcient for
the calculation of heat transfer and surface temperature. In
order to be able to calculate the temperature eld inside the
material after the surface is out of contact, the heat ow rate
Eq. (45) with a known initial wheel temperature
w0
can
be discretized for the application of the numerical solution
given in Section 3.2.
We have solved this problem with the assumption of
one-dimensional heat ow. This is justied by the consider-
ations in Section 2, except at the leading edge where on gets
a discontinuity (Fig. 6). Barber et al. [16] investigated the
heat ow at the boundary of contact areas in detail by means
of asymptotic methods. For usual operating conditions in
wheel/rail contact with very high Pclet numbers, the error
using the one-dimensional model may be neglected.
504 M. Ertz, K. Knothe / Wear 253 (2002) 498508
Fig. 6. Surface temperature of the wheel due to contact with the cold
rail, initial temperature 150

C, without frictional heating.


5. Convection
Outside the area of contact, heat ows from the wheel
into ambient air by convection at the free surfaces. With the
heat transfer coefcient , the boundary condition is

w
z
(x, z = 0) =
w
(x, z = 0) (48)
for x > a. In the case of the wheel, there is forced convection
due to the rotation. In order to nd the heat transfer coef-
cient , Fischer et al. used the empirical formula 3.6 v
with in W/Km
2
and v in m/s [11]. At v
0
= 30 m/s, one ob-
tains = 108 W/Km
2
. Another approach is possible if the
wheel is considered as a cylinder in cross ow with the uid
velocity v
0
. Using a formula from Baehr and Stephan [15],
we get = 9 W/K m
2
for the parameters given in Table 2.
As a general rule, typical values of for forced convection
with air or gas are in the range of 10100 W/Km
2
[17].
For the investigation of the convective heat transfer from
a solid into a uid, it is meaningful to consider the Biot
number
Bi =
d

w
. (49)
The main characteristic length d in wheel/rail contact is the
diameter 2a of the contact patch. With the typical values d =
2a 10 mm, = 10 W/Km
2
and
w
= 50 W/Km, we get
the Biot number Bi = 2 10
3
. At small values of the Biot
number, the temperature gradient inside the solid is small
compared with the difference of temperature between the
uid and the surface of the solid. If Bi < 0.1, the temperature
gradient in the solid can be neglected for the calculation
of the convective heat transfer [15]. The conductive heat
transfer at the surface of the wheel can, therefore, easily
be estimated with the assumption that the average wheel
Fig. 7. Surface temperature with frictional heating and convection outside
the contact patch, no initial wheel temperature, different values of heat
transfer coefcient .
temperature is constant at
w0
. This gives the heat ow rate
as
q
convection
=
w0
(50)
at the surface and the total heat ow per unit width is

Q
convection
/b

= 2r
0

w0
. (51)
With Eqs. (46) and (51)) two heat sinks exist that reduce
the temperature of the wheel. Since both terms are propor-
tional to the surface temperature of the wheel, their orders
of magnitude can easily be compared:

Q
convection

Q
rail
=
r
0

r
_

3
2av
0
. (52)
With the parameters used here, we get

Q
convection
/

Q
rail
<
1%. Convection is only taken into account on the wheel
tread, i.e. on the area that is subjected to frictional heating.
This has also been assumed by Fischer et al. [11]. The result
of Eq. (52) shows that convection can be neglected in this
case. This is conrmed by numerical calculations (Fig. 7). It
is evident that convective heat transfer can be neglected ex-
cept for values of the heat transfer coefcient that are far
from realistic. Even without convection, the surface temper-
ature outside the contact patch decreases very quickly due to
conduction from the surface into the bulk material (Fig. 3).
Convection on the sides of the wheel cannot be taken into
account with the two-dimensional model of an innitely long
cylinder. In Section 7.3, convection on the whole surface of
the real wheel will be considered approximately.
6. Steady-state wheel temperature
Since the heat loss due to conduction into the rail is pro-
portional to the initial surface temperature of the wheel, it
M. Ertz, K. Knothe / Wear 253 (2002) 498508 505
will be equal to the frictional heating at a value of the initial
temperature that can be calculated from the energy balance
P
friction
b

Q
rail
b

= 0, (53)
with P
friction
being the fraction of the frictional power dis-
sipation
P
friction
= v
s
N =

2
p
0
ab

v
s
, (54)
that ows into the wheel. Using Eq. (46), this can be solved
for the steady-state wheel temperature

=
v
s
p
0

r
_

3
a
32v
0
. (55)
The value of this temperature depends on the thermal pen-
etration coefcient of the rail only. The subscript has
been chosen to indicate that this temperature will only be
reached after a very long time at constant operating con-
ditions. While

can be calculated with little effort, the


transient calculation is much more difcult. Using a nite
element model of the wheel, the surface temperature comes
near a steady-state after 30120 min, depending on operat-
ing conditions [18]. The ratio of

to
mean
(Table 1) is

mean
=
1
0.836(1 )
_

3
32
, (56)
for the assumption of constant heat ow rate. The average
contact temperature results from the current frictional heat-
ing (Table 1) and from the average wheel temperature in
steady-state (Eq. (55)). With Eq. (44), one gets

mean,
=
mean
+

. (57)
If the heat partitioning factor is near 0.5 in Eq. (56), the
average contact temperature
mean,
in steady-state is ap-
proximately twice as high as the average contact temperature

mean
for the rst contact of the cold wheel. This is not at all
surprising. It corresponds to the commonly used assumption
that the body in continuous contact (in this case the wheel)
is an insulator [5]. Thus, all the frictional heating ows into
the rail and, therefore, doubles the contact temperature.
If we calculate the temperature eld in the wheel with
the initial temperature

from Eq. (55), it is obvious that


the temperature of the wheel remains constant at this value.
Since there is no resulting heat ow, the average bulk tem-
perature of the wheel does not change. The oscillating sur-
face temperature is due to the different distributions of the
heat ow rates resulting from frictional heating and rail con-
tact. But within a very short distance after the trailing edge
of the contact patch, the gradient in z-direction has disap-
peared and the temperature in the wheel will be constant
again (Fig. 8).
Fig. 8. Temperature under the surface for two-dimensional model, Hertzian
contact, initial wheel temperature for equilibrium without convection,
Eq. (55).
7. Application to elliptical contact areas
7.1. Frictional heating
The temperature distribution for the elliptical area of con-
tact can be calculated with the polynomial approach given in
Section 3.3. As already mentioned, the transverse heat con-
duction can be neglected and the temperature eld within
any strip parallel to the rolling direction (x-direction) can be
treated independently (Fig. 1). Since the pressure distribu-
tion within every strip is similar to that of Eq. (23), the poly-
nomial approximation Eq. (33)can be used for every strip at

i
= y
i
/b. One only has to replace a by a
i
= a
_
1
2
i
and
p
0
by p
i
= p
0
_
1
2
i
. The surface temperature is then

w
(, ) =
v
s
p
0

w
_
a
v
w
(1
2
)
0.75
F
_

_
1
2
_
,
(58)
for >
_
1
2
and || < 1 (Fig. 9). The average surface
temperature for the elliptical area of contact can also be
calculated with this approach. Using the solution within a
single strip (Table 1), and integrating this result over the
width 2b of the contact ellipse, one gets an integral that can
only be solved numerically. The accurate solution is

3D
mean
= 0.323
P
friction
b
w

av
w
. (59)
In the special case of a circular area of contact (a = b),
Archard [5] got nearly the same result with a factor of 0.31
instead. For practical applications, this result can be approx-
imated to

3D
mean

4
25
P
friction
b

av
0
, (60)
506 M. Ertz, K. Knothe / Wear 253 (2002) 498508
Fig. 9. Surface temperature for elliptical area of contact.
with the heat partitioning factor 0.5 for small sliding
velocities, i.e. v
s
v
0
, and equal thermal penetration coef-
cients of wheel and rail.
7.2. Heat conduction from wheel into rail
The heat conduction from wheel to rail due to an initial
temperature of the wheel (Section 4) can also be investigated
independently within any strip. But the calculation of the
steady-state wheel temperature is not as easy as in the two-
dimensional case (Section 6). If one assumes that the initial
temperature is constant over the whole width of the con-
tact patch, the heat ow per unit width within one strip,
Eq. (46), can be integrated over the width 2b of the con-
tact ellipse in order to get the total heat ow. This gives
approximately

Q
3D
rail
= 7b
r

w0
_
av
0
2
. (61)
The steady-state wheel temperature

without convection
can be calculated from the energy balance of frictional heat-
ing and heat loss as

3D

=
P
friction
7b
r
_
2
av
0
= 0.358
P
friction
b
r

av
0
. (62)
Another approach is possible if heat conduction between
strips is neglected. Thus, one has to calculate a steady-state
temperature for every strip as in Section 6 and take the
average value over the whole contact ellipse. This result
differs by only 4% from Eq. (62).
7.3. Convection
Convection occurs over the whole surface of the wheel.
Based on the consideration of the Biot number in Section 5,
a rst-order estimation of the heat ow due to convection is
possible. If the wheel temperature everywhere is
w0
, then
the total heat loss due to convection is

Q
3D
convection
= A
wheel

w0
, (63)
over the total wheel surface A
wheel
. Taking only the sides of
the wheel into account, one gets A
wheel
2r
2
0
. Thus, the
steady-state wheel temperature

can be calculated from


the energy balance
P
friction
+

Q
3D
rail
+ q
3D
convection
= 0. (64)
With Eqs. (61) and (63), the result is

3D

=
P
friction
7b
r

av
0
/2 +A
wheel
/
, (65)
taking into account convection as well as heat conduction
from wheel into rail. Similar to Eq. (59), the average contact
temperature in steady-state is

3D
mean,
=
3D
mean
+
3D

. (66)
It should be mentioned that the heat conduction from wheel
into rail can be calculated quite accurately compared with
convection. Since the air ow around the wheel is very com-
plicated and depends on the aerodynamics of the train, the
average heat transfer coefcient can only be estimated
roughly.
7.4. Transient behaviour
The transient thermal analysis is only possible with a nite
element model of the wheel. This has been done by Gupta
et al., but only for the frictional heating [9]. In their work,
the intermittent heat ow due to the consecutive contacts
has been replaced with an equivalent reduced heat ow rate
applied continuously to the entire circumference. The effect
of heat conduction from wheel to rail has been considered
by a modied heat partitioning factor .
With the results of Section 7.2, this approach can be
improved. Substituting the intermittent heat transfer from
wheel to rail by a continuous ow, this can be included as
a sort of convection, with the heat transfer coefcient

rail
=
7
r
r
0
_
av
0
32
3
, (67)
applied to the width 2b of the contact patch over the entire
circumference 2r
0
of the wheel. Neglecting rst the con-
vection into ambient air, this gives the same value for the
steady-state wheel temperature as in Eq. (62). On the other
hand, the transient behaviour can be calculated at any time.
Convection into ambient air can be taken into account ad-
ditionally. More details and results will be presented in a
separate paper [18].
7.5. Application to tread-braked wheels
The focus of this paper is on the contact temperatures
due to slip between wheel and rail. High temperatures and
severe thermal stresses in wheels are also caused by the
M. Ertz, K. Knothe / Wear 253 (2002) 498508 507
Table 3
Summary of results for elliptical contact area (all temperatures in

C)
Units Low speed High speed
Normal load N (kN) 100 100
Vehicle speed v
0
(m/s) 30 90
Sliding velocity (longitudinal) v
s
(m/s) 1 3
Coefcient of friction 0.3 0.1
Frictional power dissipation P
friction
(kW) 30 30 Eq. (54)
Instantaneous contact temperature
Maximal temperature
max
149.7 86.4 Eq. (37)
Average contact temperature
3D
mean
81.7 47.1 Eq. (59)
Steady-state: wheel temperature and average contact temperature
Without convection
3D

182.6 105.4 Eq. (62)

3D
mean,
173.7 100.3 Eq. (66)
= 10 W/Km
2

3D

153.5 95.0 Eq. (65)

3D
mean,
159.0 95.1 Eq. (66)
= 30 W/Km
2

3D

116.4 79.4 Eq. (65)

3D
mean,
140.3 87.2 Eq. (66)
= 100 W/Km
2

3D

63.0 50.3 Eq. (65)

3D
mean,
113.4 72.5 Eq. (66)
= 1000 W/Km
2

3D

9.1 8.8 Eq. (65)

3D
mean,
86.3 51.6 Eq. (66)
non-uniformheating which results fromtread braking [8,19].
This is not a rolling contact problem since the brake shoe
stays always in contact with the moving wheel tread and the
sliding velocity is equal to the circumferential speed of the
wheel. It can, therefore, be assumed that nearly all the heat
ows into the wheel [5]. Thus, the average contact temper-
ature between brake shoe and wheel from the current fric-
tional heating can be estimated with the equations presented
in this paper if the heat partitioning factor (Eq. (25)) is set
to unity in this case.
Some investigations of this problem neglect the periodic
contact of the wheel tread with the brake shoe and take only
the continuously increasing bulk temperature of the wheel
into account [19]. However, after long periods of braking,
the steady-state temperature of the wheel will again depend
on convection as well as on conduction from wheel to rail
through the contact patch. This can also be calculated with
Eq. (65), the frictional power now resulting from braking
and wheel/rail contact at the same time.
8. Discussion
For constant operating conditions in steady-state, the
lower limit of the average contact temperature in wheel/rail
contact is the instantaneous peak due to the current frictional
heating, i.e. without an initially increased wheel tempera-
ture. Due to heat conduction from wheel to rail, the upper
limit is approximately twice as high if convection is ne-
glected. Taking convection into account, the average contact
temperature will always be between these limit values.
In Table 3, the instantaneous contact temperature and the
steady-state wheel temperature are given for two different
vehicle speeds with equal values of the frictional power dis-
sipation. All other parameters are also equal (Table 2). The
results show that the temperature decreases with increasing
vehicle speed. The steady-state temperature

depends
mainly on heat conduction from wheel into rail. It is even
lower if convection is also taken into account. If the heat
transfer coefcient is in the range of 50100 W/Km
2
,
the heat ow into ambient air is nearly equal to the heat
conduction from wheel into rail. Thus,

is only half
as high as without convection. With the overestimation
= 1000 W/Km
2
, the average wheel temperature would
be nearly equal to ambient temperature.
The data in Tables 2 and 3 are usual operating conditions
of European locomotives. Maximum contact temperatures
in these cases are not high enough to explain thermally
induced phase transformations. This may only be the case
with extreme conditions, e.g. blocking wheels where the
sliding velocity is equal to the vehicle speed. On the other
hand, contact temperatures give rise to severe thermal
stresses. It has been shown in a number of investigations
that thermal stresses in railway wheels and rails can be in
the same order of magnitude as the stresses due to mechan-
ical loading, e.g. [8,19,20]. A surface temperature change
of 200

C would result in thermal compressive stresses

x
=
y
700 MPa. This may cause plastic deforma-
tion, residual stresses and work hardening at the surfaces of
wheel and rail. Since the thermal penetration depth is very
small, thermally induced plastic deformations are restricted
to a very thin surface layer [7,21].
508 M. Ertz, K. Knothe / Wear 253 (2002) 498508
9. Conclusions
The heat owin wheel/rail contact with smooth surfaces is
one-dimensional. Some methods for the calculation of con-
tact temperatures with this model are presented. The max-
imum surface temperature can be estimated with the ash
temperature formula of Blok. Semi-analytical and numeri-
cal methods are available for a more detailed investigation.
With a polynomial approximation, the case of Hertzian con-
tact can be investigated very efciently.
While the contact temperatures are conned to a very
thin surface layer, the bulk temperature of the wheel also
increases with time by continuous frictional heating. It can
be shown that the wheel temperature for constant operating
conditions cannot be more than twice the average temper-
ature for the rst contact of the cold wheel. This limit is
due to heat conduction from the hot wheel into the cold rail.
It corresponds to the usual assumption that the wheel is an
insulator and all the frictional heating ows into the rail.
If convection is only taken into account on the wheel
tread, i.e. on the area that is subjected to frictional heating,
it can be included into the usual model for contact tempera-
ture calculations [11]. But in this case, its inuence is small
enough to be neglected. On the other hand, real convection
occurs on the whole surface of the wheel. This can only be
considered approximately. Therefore, the steady-state tem-
perature of the wheel is even lower.
Structural changes in the rail material such as formation of
WEL are unlikely to occur due to high contact temperatures
only. They are probably the consequence of high mechanical
stresses and moderate contact temperatures that result in
high thermal stresses.
References
[1] H. Blok, Theoretical study of temperature rise at surfaces of actual
contact under oiliness lubricating conditions, Proc. Gen. Discuss.
Lubricat. Inst. Mech. Eng. London 2 (1937) 222235.
[2] J.C. Jaeger, Moving sources of heat and the temperature at sliding
contacts, Proc. R. Soc. NSW 56 (1942) 203224.
[3] H. Blok, The ash temperature concept, Wear 6 (1963) 483494.
[4] H.C. Carslaw, J.C. Jaeger, Conduction of Heat in Solids, 2nd Edition,
Oxford University Press, Oxford, 1959.
[5] J.F. Archard, The temperature of rubbing surfaces, Wear 2
(19581959) 438455.
[6] M.A. Tanvir, Temperature rise due to slip between wheel and rail
an analytical solution for Hertzian contact, Wear 61 (1980) 295308.
[7] K. Knothe, S. Liebelt, Determination of temperatures for sliding
contact with applications for wheelrail systems, Wear 189 (1995)
9199.
[8] G.J. Moyar, D.H. Stone, An analysis of the thermal contributions to
railway wheel shelling, Wear 144 (1991) 117138.
[9] V. Gupta, G.T. Hahn, P.C. Bastias, C.A. Rubin, Calculations of the
frictional heating of a locomotive wheel attending rolling plus sliding,
Wear 191 (1996) 237241.
[10] A. Cameron, A.N. Gordon, G.T. Symm, Contact temperatures in
rolling/sliding surfaces, Proc. R. Soc. A 286 (1965) 4561.
[11] F.D. Fischer, E. Werner, K. Knothe, The surface temperature of a
halfplane heated by friction and cooled by convection, Z. Angew.
Math. Mech. 81 (2001) 7581.
[12] F. Bucher, A. Theiler, K. Knothe, Normal and tangential contact
problem of surfaces with measured roughness, Wear 253 (2002)
206220.
[13] J. Gao, S.C. Lee, X. Ai, H. Nixon, An fft-based transient ash
temperature model for general three-dimensional rough surface
contacts, J. Tribol. 122 (2000) 519523.
[14] K.L. Johnson, Contact Mechanics, Cambridge University Press,
Cambridge, 1985.
[15] H.D. Baehr, K. Stephan, Heat and Mass Transfer, Springer, Berlin,
1998.
[16] J.R. Barber, F.B. Ammar, H.G. Georgiadis, Conductive heat exchange
between bodies which are in contact for a very short period of time,
Collected Papers in Heat Transfer, Vol. 123, ASME, New York,
1989, pp. 101106.
[17] W. Beitz, K.H. Kttner (Eds.), DubbelTaschenbuch fr den
Maschinenbau, 17th Edition, Springer, Berlin, 1990.
[18] J. Dohrmann, M. Ertz, K. Knothe, Transient temperature calculation
for locomotive wheels, in preparation.
[19] M.R. Johnson, R.E. Welch, K.S. Yeung, Analysis of thermal stresses
and residual stress changes in railroad wheels caused by severe drag
braking, J. Eng. Ind. 99 (1977) 1823.
[20] V. Gupta, G.T. Hahn, P.C. Bastias, C.A. Rubin, Thermal-mechanical
modelling of the rolling-plus-sliding with frictional heating of a
locomotive wheel, J. Eng. Ind. 117 (1995) 418422.
[21] G. Baumann, Untersuchungen zu Gefgestrukturen und Eigens-
chaften der Weien Schichten auf verriffelten Schienenlaufchen
(Investigations on structures and properties of white-etching layers on
corrugated rail treads), PhD Thesis, Technische Universitt, Berlin,
1998.

Vous aimerez peut-être aussi