Sei sulla pagina 1di 18

Journal of Sound and Vibration (1976) 46(3), 419-436

WHEEL/RAIL

NOISE-PART

IV : ROLLING

NOISE

P. J. REMINGTON
Bolt Beranek and Newman Inc., 50 Moulton Street,
Cambridge, Massachusetts 02138, U.S.A.
(Received 17 July 1975, and in revised form 5 January 1976)
In this paper (Part IV of a series on Wheel/Rail Noise) analytical formulas for predicting wheel/rail rolling noise in urban rail transit systems are developed based on the characterization of the wheel/rail dynamic system developed in Part I and the roughness spectra on
the running surfaces of the wheels and rails. Measured values of wheel and rail roughness
spectra taken on an operating rapid transit system are used with these formulas to predict
the rolling noise at the wayside. These predictions prove to agree well with published data on
the wayside noise from that system. In addition, measurements of the wheel roughness
spectra on a small personalized rapid transit (PRT) vehicle and the rail roughness spectra
on a section of the test track for that vehicle, when used in the analytical formulas, yield
predictions of the wayside noise during passage of the PRT vehicle that agree well with
field measurements of that noise.
1. INTRODUCIION

track
with welded steel rails, the predominant noise is rolling noise or roar. Rolling noise is
produced by the small-scale roughness on the running surfaces of the wheels and rails. This
roughness excites both the wheels and the rails; in turn, they radiate sound, producing the
characteristic broadband noise at the wayside.
In this paper, analytical formulas are presented for the response and radiation of wheels
and rails caused by the small-scale roughness on their running surfaces. These formulas draw
heavily on the analytical empirical models of the wheel-rail dynamic system described in Part
I. Wheel and rail roughness on a full-scale transit system as well as on a small-scale test vehicle
and test track were measured and, with those measurements and the analytical formulas,
the resulting rolling noise at the wayside was predicted. These formulas are validated through
comparison of the predictions with field measurements of the wayside noise.
When a transit vehicle with flanged steel wheels passes along a section of level, tangent

2. ROLLING

NOISE PREDICTIVE

FORMULAS

2.1. WHEEL/RAILEXCITATION

In an earlier paper [l], it was shown that the frequency spectrum of force, &r(a), wheel
radial velocity dbywV,(o), and rail vertical velocity a,, (CO)at the wheel/rail interface are
given by
&F(W) = (a/ I) IZ, Z&Z, + Z,) I[@w(k)+ @,&)I,

(1)

@v,v&) = @lV) lZ/(Z + Zdl [@r,(k)+ @w&)1,

(2)

@w&4 = (a/V) IZw/(Z+ ZJ I2[@r,(k)+ @,&)I,

(3)

where V is the train velocity, Z, is the wheel point impedance in the radial direction, Z, is
the rail point impedance in the vertical direction, &(k) and G,,(k) are the wavenumber
419

Downloaded from http://www.elearnica.ir

420

P. .I. REMINGTON

spectra of roughness on the rail and wheel running surfaces, respectively, k is the wavenumber,
and w is the frequency.
Though equations (l)-(3) seem to suggest that at a given frequency the force and velocity
spectra decrease with increasing vehicle speed, they do not. Instead, since k = w/V, at a
given frequency an increase in speed implies a decrease in wavenumber.
As a result, the way
the force and velocity spectra vary with train speed depends closely on the character of the
wheel and rail roughness.
At high frequencies, considerable error would result from an attempt to predict wheel and
rail velocity and interaction force by use of equations (l)-(3), with measurements
of wheel and
rail roughness spectra. The reason is that when the wheel of a rapid transit car rests on the rail,
the contact area between the two is finite in size. Consequently,
those components
of the
roughness spectrum that have wavelengths on the order of the dimensions
of the contact
patch area are effectively filtered or averaged out. This concept is discussed in more detail
below.
2.1.1.

Contact patch wavenumber filtering

When a rapid transit car wheel rests on a rail, the local elastic deformations
on the surface
of the wheel and rail produce a finite area over which the wheel and rail are in intimate
contact. The area of contact is usually shaped like an ellipse, the major and minor axes of which
depend on the loading applied to the wheel, P, the modulus of the materials in the wheel, E,,,,
and rail, E,, the radius of the wheel, a,, and the radius of curvature on the head of the rail, a,.
The lengths of the axes of the ellipse are proportional
to [2]

~/(W)Pb,a,l(a,
+ 41

Ml - v3lEJ + ((1 - vf)/-W

(4)

where v, and v, are Poisson ratios of the wheel and rail, respectively, and the expression shows
a weak dependence on all of the above quantities.
A typical transit car wheel, 30 inches
(76 cm) in diameter under 10 000 lb (44 500 N) load, resting on a rail with a radius of curvature
of 12 inches (30 cm) on the head produces a contact patch N 0.32 in x 0.26 in (8 mm x 6.6 mm).
As the running surface of the wheel passes through the contact patch and the contact patch
moves over the surface of the rail, those components of the roughness spectrum whose wavelengths are much greater than the dimensions of the contact patch are essentially unaffected.
Those components
whose wavelengths are less than or on the order of the dimension of the
contact patch may be considerably attenuated : i.e., the roughness is simply averaged out over
the area of the contact patch. In this averaging process, the wavelength along the rail is not
the only important
factor. Equally important
in determining
the degree of wavenumber
filtering is the degree of correlation
between roughness profiles measured in parallel paths
along the rail and around the wheel. If parallel paths separated by a distance on the order
of the lateral dimension of the contact patch are well correlated, the contact patch is a less
effective wavenumber filter than when those paths are poorly correlated.
Remington, Rudd and Ver [3] have presented the detailed mathematics for the derivation
of the characteristics of the contact patch wavenumber filter. It was shown that for a circular
contact patch of radius b, the filter transfer function is given by
(5)
where k is the wavenumber
along the length of the rail or around the circumference
of the
wheel, u is a constant determining
the degree of correlation
between parallel roughness
profiles at a given wavenumber, J is a Bessel function, and $ is a variable. Large a implies
poor correlation, and small c1implies strong correlation.

WHEEL/RAIL

NOISE-IV

421

: ROLLING NOISE

-50

-60

I
1

I
4

10

kb

Figure

1. Contact patch wavenumbe-r filter.

Numerical integrations of equation (5) for various values of 01are presented in Figure 1.
An examination of the curve a = 10 shows that significant filtering can occur. When wavelengths are on the order of six times the radius, b, of the contact patch (i.e., kb = l), a 10 dB
reduction in excitation level is achieved. For the contact patch size. calculated above, this
reduction would occur for wavelengths on the order of O-75in (1.9 cm) to 1 in (2.54 cm). For
a rapid transit car traveling at 40 mile/h (64 km/h), these wavelengths correspond to 750 to
1000 Hz. It should be emphasized that the derivation of equation (5) and Figure 1 requires
making some assumptions about the roughness wavenumber spectrum across the rail head
or across the wheel tread. In equation (5), it has been assumed that for a component of a
given wavenumber, k, along the length of the rail or around the circumference of the wheel,
the correlation length across the rail head or wheel head is proportional to l/ak. Other
possible assumptions are discussed in reference [3].
Combining equation (5) with equations (l)-(3), one obtains
@r&) = (02/ I) I~ZJ(Z,

+ ZJ P

lH,(k) l2[@w(k)+ @,&)I,

9, vr (4 = CmV
/ LM5 + ZJ I2IH,(k) I[@r,(k)+ @wwW>
&v,(w) = (ol v) lGJ(Z + ZJ IIH,-(k)I2[@rrW+ @wwWl,

(6)
(7)

C-3)

where k = w/V.
Upon recognition of the role of the local deformations at the wheel/rail interface in effectively filtering the short wavelength components of roughness, it was necessary immediately
to speculate on the effect of this local deformation on the wheel and rail impedances that appear
in the above response equations. This effect was examined (see the Appendix) and found to
be negligible, except at high frequency. As a result, no corrections for the contact stiffness
are included here in the response equations.
2.2. SOUNDRADIATION
The preceding response equations (7) and (8), are, of course, for the response at the point
of contact between the wheel and the rail. In what follows, these response spectra are translated into sound pressure level (SPL) spectra at the wayside.

422

P. J. REMINGTOIU

2.2.1. The wheel


Roar occurs when the wheel is excited at the tread in the radial direction. The response in
equation (7) refers to the radial motion of the wheel tread surface. However, the primary
radiating surface of the wheel (greatest area) is the web. In Part I, it was shown that r>ihration
leoels averaged over the wheel tread in the radial direction are essentialI?? equal to the axial
vibration levels averaged over the web. Thus, using equation (7) and treating the wheel as a
monopole source with uniform directivity (see Part I), one obtains the following expression
for the pressure spectrum at a distance R from the wheel :
@dw) = ~,(pcY (a2/2R) @,,;(~),
where g,,, is the radiation
impedance.

(9)

efficiency of the wheel, a is the wheel radius, and pc is the acoustic

2.2.2. The rail


For most applications,
a railroad wheel can be thought of as a point source for which the
simple model of equation (9) applies. A rail, however, can be considered as anything from a
point to a line source, depending upon the uniformity of the radiated acoustic power along the
length of the rail. If the rail is modeled as a line of incoherent sources and ifthe radiated power
per unit length is W(x), with x being the distance along the rail, then the sound pressure level
at a perpendicular
distance R from the rail can be written as
(,pq+);> = E
n

s[ I
mW(x)dx
R2

D(4),

where 4 is the angle between a line perpendicular


to the rail and the ground
is the directivity. If one assumes that the rail acts like a line of monopoles,
equation becomes

plane and D(4)


then the above

where (T, is the rail radiation efficiency, wH and ~9~are the widths of the rail head and foot,
respectively, and v,(x) is the rail velocity as a function of position. In Part I, section 3.2,
the decay of the mean square rail vibration with position was modeled approximately
as e-q
where x is the distance from the point of excitation to the point of interest. Substituting
(vf) ecqX for (v?(x)) in the above equation yields

c~(4) = MH + rF)[(Pc)~/~I(
~IR)f(mVD(4) 2):

(10)

where (vf) is the time-averaged rail velocity at the excitation point, andf(qR) = sin?RCi (qR- z/2], where Ci( ) and Si( ) are the cosine and sine integrals respectively.
The functionf(qR)
is shown in Figure 2. It is apparent that for small qR,f(qR) N 1, and the
rail is a line source. For large qR,f(qR) N l/qR, and the rail is a point source radiating from a
length of l/q. For present purposes this dependence may be approximated
by the dotted curve
in Figure 2, which isf(qR) = 1 for qR < 1 andf(qR)
= l/qR for qR > 1. Typical values of 9
for the attenuation
measured at the Pullman Standard test track as described in Part I,
section 3.2, and values of R for qR = 1 are shown in Table 1.
From equation (IO), it can be shown that the acoustic pressure spectrum, Qp(o), is given by
- cosqR[Si(qR)

@P(O) =

g,tH

WF)

where from Part I, section 5.2, D(4) N 1.

b)/~l

[_fWWRl@v,,(w),

(11)

WHEEL/RAIL NOISE-IV
IO-

f(qRI

--

-20

I
I

I
3

423

: ROLLING NOISE

I
5

- -

I
7

APPROX

MODEL

I
9

10

7R

Figure 2. Dependence of the mean square radiated pressure on the distance from the rail.

In general, for multiple excitation points such as the many wheels of a train, one simply
adds the mean square pressures produced by each, since the sources are independent. Unfortunately, equation (11) is not valid unless the point of excitation is directly opposite the
receiver: i.e., the position for measuring the acoustic pressure lies on a perpendicular to the
TABLE

Rail attenuation values based on


Pullman Standard test track data
Frequency
0-W
250
500
1000
2000

R for which qR =

tl
0.23 ft- (0.75 m-l)
0.23 ft- (0.75 m-l)
0.12 ft- (O-39m-l)
Very small
Very small

4 ft (l-2 m)
4 ft (1.2 m)
8 ft (2.4 m)
-

rail at the point of excitation. If the point of excitation actually lies a distance L down the rail,
one must distinguish two cases. The first case, L < R, can be approximately determined by
equation (11). The second case, L B R, can be derived from the exact expression for the
pressure spectrum :

which for L p R can be well approximated by


- qLePLE,(qL)] +f(qR)
where E,(qL) is the exponential integral defined by
m
E&L) =

s
il

?a.

, (12)

P. J. REMINGTOh

424

Figure 3. Distance correction for rail noise. L $ R.

The bracketed term in equation (12) is plotted in Figure 3. Multiplying the values in that
figure by eCL/R, where R is the perpendicular
distance from the rail to the receiver, gives the
appropriate
distance correction for the SPL radiated by the rail. Figure 3 shows a strong
tendency at large QC for the SPL from the rail to remain constant with distance from the rail.
The reason for this is that the contribution
to the rail noise is primarily from the rail vibration
at the excitation point, and for L B R the distance from the excitation point to the receiver
does not change significantly
as one moves away from the rail. For small q,C, curves in
Figure 3 are nearly flat in qR, implying a l/R dependence of the rail radiation with distance,
because most of the radiation comes from that portion of the rail directly opposite the
receiver. Also, because the vibration decays slowly along the length of the rail, the rail appears
to be a line source reduced in strength by emVL.
By using equation (11) or (12), one can calculate the SPL from the rail due to a single
excitation point. The SPL of many excitation points is the sum of the mean square pressure
from each.
2.2.3. Predictive formulas
Equation (8), appropriately
combined with equation (11) or (12), and equation (7), combined with equation (9), yield an analytical prediction of the pressure spectrum produced by a
single wheel interacting with a rail. It is useful to rewrite these expressions in terms of the SPL
in a given frequency band, Aw, by dividing the pressure spectrum by pg(2 x low5 N/m2),
where the parenthetical
term is the standard reference pressure, multiplying both sides of the
equation by Aw, and taking 10 times the logarithm of both sides. For the rail, this rewriting
yields
SPLI;1, = lOloga,

+ lOlog [(wH + w+)/nR] + IOlog IZ,/(Z,

+ lOlog lHf(k)12 + lOlog[(pc~/p,)~

+ Z,)l +

Q,,(k) Ak] + lOlogG(qR,qL),

where Ak = do/V, Q,,(k) is the sum of the wheel and rail roughness wavenumber
V is the train speed, and G(qR, qL) is given as G(qR, qL) = f(qR), for L -g R, and
G(?R, vL) = e-?(qR/r&)

cash &(l

- @) enL E,W)

+ f(nR)l,

(13)
spectra,

WHEEL/RAIL NOISE-IV

425

: ROLLING NOISE

for L $=R, where R is the perpendicular distance from the rail to the receiver and L is the
distance along the rail from that perpendicular to the excitation point. For the wheel,
SPL\z = lOloge,

+ 1010g(a2/2Rr2) + lOlog ]Z,/(Z, +Z,)]+

+ 10 log

IH,(k) I2+ 10log KPc~/P,)2@w,(k)


N,

(14)

where R is the distance from the wheel to the receiver.


Equations (13) and (14) constitute the formulas for predicting rolling noise due to wheel/rail
interaction. They require knowledge of the radiation efficiency, 0, of wheels and rails, the
geometry (wheel radius, distance of the wheel and rail to the receiver, rail head-and-foot
widths), the wheel and rail impedances Z, and Z,, the geometry of the contact patch (which
depends on wheel/rail geometry and load), and the roughness spectrum on the wheels and
rails. Until recently no measurements of @,,(k) existed. In Part V of this series of papers
a device built to obtain the required roughness data is described.
2.3.

SAMPLE EVALUATION OF ROLLING NOISE

Given data on the roughness of wheels and rails and the analytical empirical models of the
wheel/rail dynamic system described in Part I, one can estimate the rolling noise at the wayside caused by the passage of a transit vehicle. As a preliminary check on the predictive
formulas, wheel and rail roughness data measured on the Massachusetts Bay Transit
Authority (MBTA) were combined with equations (13) and (14) to make an order-of-magnitude comparison of our rolling noise estimates with the data of Rickley and Quinn [4] taken
on the South Shore line of the MBTA (welded rail) just after that line was opened. Figure 4
shows the roughness wavenumber spectrum measured on, first, a standard 24.5 inch (62 cm)
revenue-service MBTA steel wheel in good condition and, second, a rail of the Green Line
WAVENUMBER

(rad/cm)

--60

-1201

1
04

1111111

1 11

IIIII

I.0
WAVENUMBER

11

IO
(RADIANS/in

llul
ID0

Figure 4. MBTA wheel and rail roughness.

426

P. J. REMINGTOh

of the MBTA near the Forest Hills Station. These data and the device used to measure them
are described in Part V.
In what follows, calculations are described of the SPL at 25 ft (7.6 m) from the track for a
train composed of four MBTA Type I South Shore Rapid Transit cars. The cars are 70 ft
long (21.4 m), have trucks with a 6.83 ft (- 2.1 m) wheelbase, have a center-to-center
truck
length of 51 ft (15.6 m), use 28 inch (0.71 m) diameter wheels, and weigh 5 60 000 lb (27 200
kg). The rail and wheel roughness data in Figure 4 is used and a 50 mile/h (80 km/h) train speed.
These specifications correspond
to a number of measurements
performed by Rickley and
Quinn [4].
2.3.1. Typical results
MBTA rail is typically AREA 115 lb/yd (56 kg/m) and the wheels have N 1000 lb (454 kg)
mass with a head cross-section
5.5 inch wide by 2.75 inch thick (13.9 x 6.9 cm). From this
information
and the models in Part I, one can estimate (i) the rail impedance (equation (2) in
Part I), (ii) the wheel impedance (equations (3) and (4) in Part I), (iii) the rail radiation
efficiency (equations (6) and (7) in Part I), and (iv) the wheel radiation efficiency (equation (8)
in Part I), all of which are required in equations (13) and (14).
Taking the load per wheel to be lO,OOO/lb (44,500 N), one obtains a contact patch of 0.32 x
0.26 in (8 mm x 6.5 mm). In order to use the filter characteristics
of Figure 1, one can use an
equivalent radius defined by b = +Vcd= 0.15 in (6.8 mm), where c and dare the length of the
major and minor axes of the elliptical contact patch. This relation is based simply on equivalent areas in the circle and ellipse.
When one sums the contributions
from the 16 wheels on the train, the wheels and rail on
the opposite side of the track may be ignored, on the assumption that the car body effectively
blocks their contribution.
(Underprediction
will be, at most, only 3 dB.) The geometry of the
situation when the train is opposite the observer is shown in Figure 5. Only the four wheels
directly opposite the observer are significant wheel sources. However. all 16 wheels excite the
rail, and one must account for each. At some frequencies, when the loss factor, q, which
determines the decay rate along the rail, is large, the excitation of the rail by a wheel far down
the track will not contribute significantly to noise perceived by the observer. However, if q is
small, the excitation produced by a wheel far down the track will result in significant vibration
in the rail directly opposite the observer. The distance for sound propagation
from the rail
is then R, as opposed to R (see Figure 5) and, as a result, the excitation in the rail produced
by the wheel may be a significant source of noise.
Using equations (13) and (14) and the data in Table 1, one finds that below 2000 Hz only
the excitation produced in the rail by the four wheels directly opposite the observer contributes significantly to the radiated noise. Above 2000 Hz, the excitation produced by all

C42m-

Figure 5. Four-car train showing the distance parallel to the track in meters from the observer toeach wheel.

WHEEL/RAIL NOISE-IV

aB

427

: ROLLING NOISE

60

60

50
"
IO

"
20

"
40

"
60

"
160

'1
315

FREOUENCY

"
630

"
1250

"
2500

"
5000

10000

Iii11

Figure 6. Comparison of rolling noise predictions and measured data for a four-car MBTA train running
at 50 mile/h (80 km/h) and measured at 25 ft (7.5 m).

16 wheels is significant. (Note that this is an approximation based on quite limited measurements of r~.Extrapolating measurements made over a 5 m length of track to 40 m could lead
to errors but, at most, one is introducing only a 6 dB overestimation of the SPL.)
Figure 6 shows the result of carrying out the above estimations. The figure presents both the
wheel and rail contributions and their sum, and compares them with the range of SPL levels
measured during five passbys. The predictions fall acceptably within the range of measured
data. The wheel contribution dominates only at low frequencies, while the rail contribution
dominates over the rest of thefrequency range. However, this result should not be taken as one
of general applicability, since the relative noise contributions of wheel and rail depend on
both the distance of the observer? from the rail and the relative magnitude of the wheel and
rail impedance.
3. VERIFICATION OF PREDICTION FORMULAS FOR ROLLING NOISE
Rolling noise was measured at the Champ Carry Technical Center test track of Pullman
Standard (P-S) in Hammond, Indiana. These measurements can be compared with predictions based on the analytical/empirical formulas given in Part I and the previous section.
3.1. DESCRIPTION OF P-S TESTS
At P-S, we measured the noise produced by the interaction of the steel wheels of a small
PRT (personalized rapid transit) vehicle with the rails of the test track. Ideally, we would have
performed measurements on a full-scale transit system, but the lower cost and simplicity of
operation of the PRT system at P-S led us to perform the verification measurements there.
In addition, the flanged steel wheels of the PRT vehicle and the steel rails of the test track are
basically the same, except for size, as those of a full-scale transit system. Consequently, if the
predictions agree with the measurements of noise and vibration from the PRT vehicle on the
test track, it is believed that they will also agree with measurements on full-scale transit
systems.
t The wheels of the train can be characterized as a line of point sources, and the rail can be characterized as
either a line source or a point source, depending on the observers location and the rate of decay of vibration
along the rail, q (see Part I). As a result, the sound radiated by the wheels and that radiated by the rail decay
with distance from the track at different rates, depending on geometry, observer location, and t].

428
3.1.1.

P.J.REMINGTOI\I
ThePRTvehicle

The small PRT vehicle (Figure 7) was originally fabricated as an engineering


test model.
The PRT has four 14 in (35.5 cm) diameter 140 lb (63.5 kg) steel wheels, each of which is
individually
powered by a hydraulic motor. The forward and aft bolsters (metal frames

Figure 7. PRT (Personalized Rapid Transit) vehicle.


containing the two hydraulic motors and the associated wheel bearings) are each suspended
independently
on four coil springs with hydraulic dampers. Specifications
for the vehicle
are given in Table 2.
TABLE 2
Vehicle specij7cations

Vehicle weight
Wheel base
Width over side sills
Length over end sills
Height of the floor above the rail
Extreme height of vehicle

6500 lb (2950 kg)


8 ft (2.43 m)
4.5 ft (1.38 m)
12 ft (3.66 m)
- 2 ft (0.61 m)
- 7 ft (2.13 m)

3.1.2. The test track


The test track is an oval, approximately
l/3 mile (0.53 km) around. The rail, which consists
of both welded and bolted ASCE 60 lb/yd (30 kg/m) steel rail at 42 in (I.07 m) gauge, is laid
on wood ties 4 in x 6 in x 6 ft (10 cm x 15.2 cm x 1.83 m) on crushed limestone ballast.
A three-phase power rail is located on the inner side of the oval. A section of straight, level,
welded rail approximately
10 ft (3-l m) long was selected as a test section for rolling noise.
Wayside noise, bolster vibration, and rail vibration measurements
were taken as the PRT
vehicle passed through this section at various speeds. Prior to these tests, the roughness of
the wheel and rail running surfaces was measured, by using the device described in Part V.

WHEEL/RAIL NOISE-IV

: ROLLING NOISE

429

3.1.3. Roughness measurements


The roughness of two parallel strips, or lines of contact between the wheel and the rail, were
measured on the outside rail. The 1,/3-octave band wavenumber spectra of the two strips are
shown in Figure 8. There is considerable difference in roughness at high wavenumbers between
them. Since it is not always possible to know exactly where the wheel contacts the rail, this
spread in rail roughness will lead to a band of uncertainty in our predictions of rolling noise at
high frequency.
WAVENUMBER
0.1

-40

(rod

/cm

IO

-40

-50

- 120

0.1

I111111

Illlll

1-o
WAVENUMBER

10
(rad

/ I

I IIIILJ
lC0

Figure 8. Rail roughness on the Pullman Standard test track.

In addition to measuring rail roughness, we also measured the roughness of one of the
wheels. The wheel was mounted in a lathe and measured for roughness before and after it was
turned smooth. Figure 9 shows the roughness spectra before and after turning. The turned
wheel is considerably smoother; in fact, it is so much smoother than the rail that for all
practical purposes its roughness can be ignored.
3.1.4. Instrumentation
As the vehicle passed through the test section, noise was measured by a microphone
(+ in (1.25 cm), Brtiel and Kjaer 4133) with a windscreen hung on a boom suspended from the
vehicle so that the microphone was 3 ft (0.91 m) from the face of the wheel on the axis of the
wheel. The microphone always hung on the opposite side of the vehicle from the power rail.
Both bolster and rail acceleration were also measured. Although only noise data are presented
here, the interested reader will find the acceleration data reported in reference [3]. All the
data were recorded on Kudelski 2-track NAGRA IV SJ tape recorders for later analysis.

430

P. J. REMINGTON
WAVENUMBER

,171, CT

Figure 9. Roughness spectra of 14 in (35.5 cm) diameter PRT standard wheels.

For the noise and vibration tests, the PRT vehicle was driven over the test section in both
directions at 10, 20, and 25 mile/h (16, 32 and 40 km/h).? A mark was placed on the voice
tracks of the tape recorders as the vehicle reached the center of the test section. In the
laboratory,
the data was reduced to l/3-octave band spectra by passing the recorded signals
through a General Radio real-time analyzer. A comparison
of these data with predictions
is made in the following sections.
3.2. COMPARISON OF ANALYTICAL PREDICTIONS AND MEASURED DATA
Using the wheel/rail dynamic system models of Part I in conjunction
with equations (13)
and (14), one can predict the SPL at the wayside caused by passage of the PRT vehicle through
the test section and compare it with the measured data. This SPL is composed of radiation
from the rail excited by each of two wheels on one side of the vehicle and radiation from each
of these two wheels. Note that here radiation from the wheels and rails on the opposite side of
the vehicle is neglected. Because the microphone is so close to the wheels and rails on one side,
no more than a 1 or 2 dB error in the predictions should result.
For the ASCE 60 Ib/yd (30 kg/m) rail of the P-S test track, the width of the head is 2+ in
(6 cm) and the width of the foot is 44 in (10.8 cm). The wheels are 14 in (0.35 m) in diameter.
These parameters yield the following expressions: namely, for the SPL radiated by the rail
from excitation by a single wheel,
SPL,=-12*5+

lOloga,+

lolog~z,/(z,+z,)~~+

+ lOlog [(~cc@oY @r,(k) dkl

t Previous

measurements

IOlogIH,(k)/2+

+ 10log GO,%tIL),

(15)

[5] were made of the noise of the PRT vehicle when it was running, but when the
wheels were raised above the ground. They indicated that propulsion motor noise under those conditions was
15 to 20 dB(A) lower than the noise produced when the vehicle was run on the test track rail at lo,20 and 25
mile/h (16,32 and 40 km/h). There is strong indication, therefore, that wheel/rail noise is the dominant source
at these speeds.

WHEEL/RAIL NOISE-IV

431

: ROLLING NOISE

and, for the SPL radiated by a single wheel,


sPL,=+lolog~z,/(zW+z,)~Z+

lOlog~Hf(k)(2 +

+ 10 log [(pco/~0)2 @,,(k)

+ 20h3 WR),

(16)

where R is the distance from the observer to the wheel of interest.


The wheels each weigh N 140 lb (64 kg) and have a tread cross-section (apart from the
flange) of 4 in x 1 in (10 cm x 2.5 cm). The 60 lb/yd (30 kg/m) rail has a radius of gyration of
1.57 in (4 cm). Using this information and the analytical formulas of Part I, one can estimate
(as was done for MBTA wheels and rails in section 2.3) the wheel and rail impedances and
radiation efficiencies required in equations (15) and (16).
The loaded weight of the PRT vehicle is N 7000 lb (3180 kg), which givesa 1750 lb (77OON)
load applied to each wheel. The radius of curvature of the rail head is 12 in (O-31m). The contact patch, then, is an ellipse with major and minor axes of length O-12in x 0.23 in (3 mm x 5.9
mm). As before, this may be approximated by a circle having a radius b of b = +1/(0*12)(0*23)
= 0.083 in (2.1 mm). The contact patch wavenumber filter can then be estimated from
Figure 1. Using the estimate for CL= 10 in Figure 1 for the contact patch wavenumber filter
and the rail roughness data of Figure 4 (note that the turned wheels are much less rough than
the rails), one can calculate the excitation applied to the wheels and rails as a function of
frequency at lo,20 and 25 mph (16, 32 and 40 km/h), by recalling that k = w/V.
The wheelbase of the PRT vehicle is 8 ft (2.45 m). From equation (16), it can be shown that
only the radiation from the wheel opposite the microphone is significant; from equation (15),
it can be shown that below 2000 Hz only the excitation in the rail produced by that wheel is
significant in causing the rail to radiate sound. Above 2000 Hz, both wheels cause the rail to
radiate equally.
Comparisons of measured and predicted radiated sound are shown in Figures 10 through
12. The spread in the predicted noise levels is due to the spread in the roughness spectra measured
in parallel strips on the rail as shown in Figure 8. In general, the agreement is quite good, except
that at low frequencies there is a tendency to underpredict the radiated sound. If the effect of
the Hertzian contact stiffness on the wheel and rail impedance had been included, these lowfrequency levels would have been raised approximately 3 dB [2], certainly improving the

40

IO

IIl
I
20

40

80

160

315

FREOUENCY

630

1250

2500

5000

10000

(Hz 1

Figure 10. Wheel/rail noise at 25 mile/h (40 km/h) at 3 ft (0.91 m) from the wheel.

432

P. J. REMINGTON

Zj

50-

PREDICTED
- WHEEL SPL
. RAIL SPL

k?

FREOUENCY

I H: !

Figure 11. Wheel/rail noise at 20 mile/h (32 km/h) at 3 ft (0.91 m) from the wheel.

agreement. However, within the limits of the approximate


models used to develop the predictions, the agreement is as good as could be expected. The major conclusions to be drawn
from the results are that for this case the wheeldominates the soundradiation at lowfrequencies
(< 315 Hz) and high frequencies (< 2000 Hz), and the rail dominates in the mid-frequencies
(400 Hz to 1600 Hz). The physical explanation
for these observations
is as follows. At low
frequencies, the wheel impedance is only slightly larger than the rail impedance, implying that
response levels are comparable.
However, the rail is an inefficient radiator at these low frequencies; thus, wheel radiation dominates. In the mid-frequencies,
the rail impedance is much
less than the wheel impedance, implying higher rail response levels; since the rail is an efficient

40111~1111
40
10
20

80

1lIII

1
160

315

630

11
1250

11
2500

11
5000

10000

FREOUENCY
(HZ1
Figure 12. Wheel/rail noise at 10 mile/h (16 km/h) at 3 ft (0.91 m) from the wheel.

WHEEL/RAIL

NOISE-IV

: ROLLING

NOISE

433

radiator, the sound radiation from the rail dominates. At high frequencies, the wheel impedance drops below the rail impedance, resulting in higher wheel response and, hence,
higher sound radiation 1evels.t
4. CONCLUSIONS
Analytical formulas for the prediction of rolling noise (i.e., the noise produced by the smallscale roughness on wheels and rails) from steel-wheeled rapid transit vehicles have been
developed and verified through comparison of their predictions with field measurements both
from a full-scale transit system as well as a small-scale engineering test vehicle.
The size of the contact patch at the wheel/rail interface has been shown to have a significant
effect on the magnitude of the noise radiated, although the effect of the Hertzian contact
stiffness itself on the wheel and rail impedances has not been found to have a significant effect
on the radiated noise.
For the two cases examined, the MBTA and the PRT system at P-S, the wheel has been
found to be the dominant source of noise at low frequencies, primarily because at those
frequencies, the rail is an inefficient radiator. In the mid-frequencies, the rail has been seen
to be the dominant radiator, primarily because there the rail impedance is sufficiently lower
than the wheel impedance to imply that the response of, and hence the sound radiation from,
the rail is greater than that from the wheel. At high frequencies, either the wheel or the rail
may dominate the radiation, depending on the location of the receiver and the relative magnitudes of the wheel and rail impedances.
The earlier results of Bender and Remington [l] agree only in part with these conclusions.
They found that the rail was a significant contributor to wayside noise only in the 400-800 Hz
and 4-6.3 kHz frequency ranges; in other frequency ranges (especially at the peak in the wayside noise spectrum at 1 to 1.6 kHz), they found the rail contribution to be on the order of
10 dB below the measured wayside noise. In the investigation reported in this paper, however,
it was found that (1) the rail is the major contributor over a broader frequency range, and
(2) more importantly, when the wheel/rail noise spectrum peaks, the rail is the dominant
contributor, as illustrated in Figures 6 and 10-12. We suggest that other sources, such as
electric motors and gears in the drive train, are responsible for the high wayside noise levels
in the data of Bender and Remington.
ACKNOWLEDGMENTS
The study reported here was supported under contract from the U.S. Department of
Transportation. The author is grateful for this support and for the advice and help of Dr
Robert Lotz of the Department of Transportation, Drs Erich K. Bender and Michael J. Rudd
of Bolt Beranek and Newman, Inc,; and Mr Harold Gramse of Pullman Standard. The
author is also grateful to Dr Larry Wittig and Mr Mark Myles of Bolt Beranek and Newman,
Inc. for the care they exercised in obtaining the field measurements reported here.
REFERENCES
1. E. K. BENDER
and P. J. REMINGTON1974Journal of Soundand Vibration 37,321-334. The influence

of rails on train noise.


2. S. TIMOSHENKO
and J. N. GIDDIER 1951 Theory of Elasticity. New York: McGraw-Hill Book
Company Inc. See Chapter 13, Article 126.
t Similar comparisons of prediction and measurement have been carried out for resilient and damped
wheels installed on the P-S PRT vehicle with comparable agreement. For the sake of brevity, these are not
included here. Reference [3] gives more detail.

434

I. J.REMINGTC)\

3. P. J.

REMINGTON,
M. J. Ruon and I. L. V~R 1975 U.S. Department ojTransportation Report No.
UMTA-MA-06-0025-75-10.
Wheel/rail noise and vibration.
4. E. J. RICKLEY and R. W. QUINN 1972 U.S. Departmcrlt ofTransportation Report No. DOT-TSCOST-72-31. MBTA rapid transit system (Red Line, wayside and in-car noise and vibration
measurement.
5. H. E. GRAMSE and J. H. SPENCER 1974 U.S. Departmetlt o/Transportatiotl Report No. UMTA-hiA06-0027-74-I. Noise and vibration of a steel wheel/steel rail personalized rapid transit system.

APPENDIX:

THE

EFFECTS

OF LOCAL

If the local Hertzian


contact
stiffness
effective impedance of the wheel and rail
One can examine the possibility in detail
approximate model of Figure Al. In that

DEFORMATION

ON RESPONSE

of the wheel in contact with the rail is small, the


could be reduced and, thus, the response affected.
by linearizing the contact stiffness and using the
figure, the contact stiffness, Kc, is split between the

i,, (WHEEL IMPEDANCEI

Figure Al. Simple contact stiffness model.

wheel and the rail, and the motions on the surface of the wheel and rail are each split into two
parts: those caused by the contact stiffness, V,, and V,,, and those caused by the impedances,
VWand V,. (This model implies that the impedances have been measured under circumstances
where the impedance of the contact stiffness is much greater than Z, or Z,.) The latter motions
are those that result in sound radiation from the wheel and rail, since the motions at the
contact point from the contact stiffness are very localized.
Solving for V, and V,, one obtains

2zv-w-G)1 ~ro*hncss
(

VW= [Z&Z,

+ Z, +

vr = [UG,

+ -TV+ =zv/zc)1

~roughness,

(Al)
(A2)

is the roughness velocity from the wheel and rail sensed as


where Z, = 2KJjw and VrOughneSS
the wheels roll over the rail. Clearly, if Z, g Z, and Z, 9 Z,, then the above equations simplify
to
VW(O) = [Z,/(Z,. +

Zdl [f(w) + k(w)17

V,(w) = [Z,/(Z, + Z,)] [i(o) + Mo)l.


Equations

(6) through

(8) in the text are unaffected

by the contact

stiffness.

WHEEL/RAIL

NOISE-IV

: ROLLING NOISE

435

The author knows of no formula for calculating KCfor two cylinders in contact, i.e., a wheel
and a rail; however, one can estimate KC for an elastic sphere pressed into a half-plane as
KC = 1*22(E2aP)J3, where E is the modulus of the material, a is the sphere radius, and P is
the load applied to the sphere. Simulating the parameters of the PRT vehicle and wheel, for
a 14 in (355 cm) diameter sphere (to simulate a wheel of similar radius) under 1750 lb (7800 N)
load, one obtains KC = 3.3 x 10 lb/ft (4.8 x lo* N/m). In Figure A2 this contact impedance
is compared to results from the analytical impedance models developed in Part I for the PRT
wheel and the Pullman Standard test track rail (ASCE 60 lb/yd (30 kg/m)). Above N 500 Hz,
the contact impedance becomes comparable to or less than the wheel impedance, and some
effect would be expected. Below 500 Hz, little effect would be expected.
IO

IO'

IO'
=:

-z

Y
v)

D
;

lOa

IO'

>

16
IO5

IO'
IO

IO2
IO

20

40

80

315

160

630

FREWENCY

Figure A2. Comparison

1250

2500

so00

10000

(Hz)

of the contact impedance with the wheel and rail impedance.

It is useful to distinguish two regions in Figure A2: the region where 2, 9 2, and the
region whereZ, B Z,. In those two regions, it is possible to simplify equations (Al) and (A2),
ar shown in Table Al. In Region I the contact stiffness modifies the wheel and rail response by
Z&Z, + 25), and in Region II the responses are modified by Z,/(Z, + 22,). Consequently,
to examine the effects of the contact stiffness, one need only examine Z&Z, + 25) in Region
I and Z,/(Z, + 22,) in Region II. Using the analytical models of Part I, one finds that
Z, z 103(f/1000)2(1 -j)
Z, = 4 x 10(f/1000)(1
Z, = j9.90 x 106/f

(lb s/ft), (Region II)


-j)

(lb s/ft) (Region I)

(lb s/ft). (Regions I and II)

Figure A3 shows the resulting effects of the contact stiffness in the two regions.
TABLE

Al

Wheel and rail response including


contact stiflness

436

Figure A3. Effect of contact stiffness.

Previous calculations
these calculations

were performed
Z,/(Z,

on the assumption

+ 22,) = Z&Z,

+ 22,)

that Z, + cc. As a result, in all


= 1,

and Figure A3 shows the correction factor that should be added to all the rolling noise predictions to correct for the contact stiffness. Only at very high frequency (> 4000 Hz), where
Z, is much less than both Z, and Z,, is this correction significant.

Potrebbero piacerti anche