VEHICLE/TRACK

DYNAMICS AND
THE WHEEL-RAIL
INTERFACE

ROLLING RESISTANCE REVISITED

Frank Szanto HOST SPONSORS

Downer Rail
PLATINUM PARTNER
R
Corresponding Author: [email protected] Broc

SUPPORTED BY

SUMMARY
Resistance to motion is an important input for train performance simulations, and the Davis equation is a
well-known resistance formula, widely used for freight trains. It is known to give conservative values, and
has numerous variants such as Modified Davis and Adjusted Davis with factors of e.g. 0.7 being applied to
the formula. There is confusion about the right constants to use, and frequently specifications for equipment
contain values which are inappropriate.

This paper begins with a review of some of the published formulae, and then presents a methodology for
calculating the overall resistance of trains of known weight and composition operating on known track
geometry, using tractive effort, GPS location and speed records downloaded from locomotive data loggers.
A simulation program was modified to take the time history of tractive effort as an input, and to calculate
velocity, which is then compared with the recorded speed. The “Davis” formula parameters are adjusted
iteratively until the calculated speed approximately matches the actual speed. This approach is used to
benchmark values of published constants and to suggest appropriate values for trains operating in Australia.

1. INTRODUCTION The “Modified” Davis equation for Freight Cars

published in AAR RP-548 [2] takes the form:
Train performance simulations are used for
planning new track alignments, for predicting the
R’ = 1.3 + 72.5/wn + 0.015v + 0.055v2/wn (2)
maximum loads which can be hauled, or the speed
or fuel consumption for proposed train
Note the second term is a constant by the number
configurations on existing track. Resistance to
of axles. In Davis’ work, when most rollingstock
motion is an important input for any such
had journal bearings, it was 29lb/axle. The
simulation, and the Davis equation, which is a well-
constant in RP-548 is based on 18lb/axle, for a
known resistance formula, is widely used for freight
four axle wagon with roller bearings, which is not
trains.
immediately obvious in the formula above.
The Davis equation is a quadratic in velocity of the
2 In RP-548, the last (aerodynamic drag) term
form R = A + Bv + Cv , with constant coefficients
incorporates a cross-sectional area of 110 sq.ft.
based on experimental work done by Davis in 1926

R
Other variants of the equation have drag
[1]. It is known to give conservative resistance
coefficients according to wagon type.
values, and the formula has numerous variants
such as Modified Davis and Adjusted Davis. There
Converting to metric units, (with g= 10 m/s2):

Broc
is confusion about the right constants to use.
.
R’ = 6.5 + 320/wn + 0.046v + 0.096 v2/wn N/t (3)
2. DAVIS EQUATION
Note that these constants have an accuracy of less
The original form of the Davis equation for freight than two decimal places, hence the use of 10 m/s2
cars was: in the conversions for convenience..
R’ = 1.3 + 29/w + 0.045v + 0.0005av2/wn (1)
3. COMPONENTS IN THE EQUATION

R
where R’ is resistance in lb/ton
Table 1 following compares the contribution of
w is axle load in tons
each component of resistance, for a wagon when
(here ton= short ton = 2000 lb)

Broc
loaded (80 tonne gross) and empty (20 tonne tare).
n is the number of axles
RTED BY R is the total resistance, and R’ the Davis
a is the frontal area in sq.ft
resistance in N/tonne.
Note my (non-standard) use of R’, to distinguish
.
per unit resistance from total R, in units of force.

SUPPORTED BY
 Frank Szanto ROLLING RESISTANCE REVISITED
Downer Rail

V (km/h) Mass (t) A A0 B C R (N) R’(N/t)

20 80 520 320 73.6 37.8 951 11.9
20 130 320 18.4 37.8 506 25.3
80 80 520 320 294 605 1740 21.7
20 130 320 73.6 606 1128 56.4
100 80 520 320 368 946 2154 26.9
20 130 320 92 946 1488 74.4
Table 1 Resistance components for 80 tonne wagon as per AAR RP-548

At 20 km/h, the rolling friction (A+ A0) terms make We were not alone in finding this anomalous result.
up 88% of the total resistance for the wagon at A 2006 paper by Kim, Kwon, Kim and Park [4]
gross, and 47% at tare. The minimum continuous describes resistance values derived from coasting
speed of a locomotive is typically around 20 km/h, tests during the introduction of the KTX high speed
so only the A and A0 terms come into play when a train in Korea. The initial result found a negative
loaded train crawls up a steep grade. At higher value for B, of -19.5 N/km/h for a train weighing
speeds, the C (air drag) term may be more than 326 tonnes. The tests were intended to compare
half of the total resistance. the aerodynamic drag of the KTX with the TGV,
from which it was derived. The Koreans assumed
The A terms give resistance equivalent to a 0.12% the French coefficients were authoritative, and re-
or 1/824 grade for the loaded wagon, and 0.26% or analysed their data accordingly, which changed the
1/388 for the empty wagon above. Small variations values determined for the air drag coefficient.
in grade may therefore cause significant errors
when trying to measure resistance. Forces proportional to velocity are associated with
damping, or laminar flow. The B term is often
Calculating the resistance R’ in the units of the A described as “flanging”, but unless a vehicle is
term (lb/ton) can be confusing as adding mass hunting badly, the flanges do not make contact on
reduces unit resistance, since the air resistance tangent track. Even if they did, sliding friction would
term is independent of axle load. It is only useful not give resistance increasing linearly with velocity.
for calculations using equivalent grades. For the In fact, Coulomb friction is high at near zero
cases in this table, the total resistance R varies by velocity, and then decreases, which might explain
a factor of 4, while R’ as calculated by Davis varies a negative value. Other “dynamic” effects such as
by a factor of 6, and yet the value (26.9) for the draft gear action are considered to be covered by
loaded wagon at 100 km/h is almost the same this term, but there seems to be no theoretical
(25.3) as the empty wagon at 20 km/h. R’ therefore basis for this. Based on the above, I believe the
provides little insight into the train’s behaviour. It is velocity coefficient is very small, and can be taken
clearer to write the equation for the resistance per as zero.
wagon of mass m and n axles as:
5. OTHER FORMULATIONS
R= 6.5m + 80n + 0.046vm + 0.096 v2 (N) (4)
In 1990, Canadian National (quoted in [3])
4. IS B BOGUS?
published a version of the Davis formula with A =
It is questionable whether the B term has any 1.5 lb/ton. AAR also suggested that A can vary
validity. As mentioned in the AREMA Manual [3], from 0.7 to 2.13 lb/ton [3]. There are many factors
some work by AAR in 1988 suggested this term which affect the “constant”.
should be zero, and I have seen evidence to
support this. In 2000, when the 4000 Class A different approach is often used in passenger
locomotive was introduced in Queensland, some train formulae for the velocity independent term,
tests to determine rolling resistance were with the product of mass and number of axles n
performed by measuring deceleration while under a square root function. For example, SNCF
coasting to rest, and curve fitting the data. The uses the following formula for double deck EMUs

R
constant term was found to be 1262N, which was
close to the AAR formula (albeit for a wagon). The
R  14  10  n  M (N)
Where n is number of axles
(5)

Broc
air drag term also agreed reasonably well with the
RP548 value, but oddly, the B term was negative, M is mass in tonnes
i.e -265 N/km/h. It was assumed this was a
mistake. Surprisingly, this non-linear function gives values
which are not significantly different from Davis, as
shown below in Table 2 for the 80 tonne wagon
case:

SUPPORTED BY
 Frank Szanto ROLLING RESISTANCE REVISITED
Downer Rail

independent of mass and should be calculated

separately.
Case Mass SCNF Davis Diff
Loaded 80 T 792 N 840 N 6% The drag value calculated in the coasting tests for
Half 50 T 626 N 645 N 3% the 4000 class locomotive was 0.24v2 [N]. Using a
Empty 20 T 396 N 450 N 12% cross-sectional area of 11 m2 with the RP-548
Table 2 Resistance as per Davis vs SCNF value of 0.0024 in the table gives 0.54 v2 which is
at bit more than double. However, the following
assumes the AAR value for the locomotive, and
While my initial intention was to use experimental only considers wagon resistance.
data to check which approach is more valid, the
small differences in the results might preclude this.
Nevertheless, both formulas will give double the 7. EXPERIMENTAL COMPARISONS
resistance if the number of wagons (and hence
axles) is doubled. Being linear, Davis is the simpler The following section uses data obtained from
formulation, which passes Ockham’s Razor. locomotive event recorders to benchmark Davis
formula results. Note that these were not special,
instrumented runs, but trains in normal service. It
6. AERODYNAMIC DRAG would be more accurate to measure resistance by
As above, the formula in RP-548 has a “one size hauling trains on flat track, and measuring the
fits all” approach, and combines an area of 110sqft forces in the coupler of the lead wagon using strain
with a drag coefficient of 0.0005 into one term, but gauges, but this has the disadvantage that it
there are other published values for various requires a great deal of preparation and it is hard
wagons. Table 3 following compares the RP-548 to find perfectly flat track of sufficient length. The
value with some values published by Canadian method following tries to make use of data which is
National [3]: easily obtained, but this necessarily includes many
variables such as wind conditions, which must be
Coeff A sq.ft CxA Metric accounted for.
AAR 0.0005 110 0.055 0.096 Event recorder data typically includes tractive
RP548 effort, speed and GPS data for the locomotive,
Coal 0.00042 105 0.044 0.077 recorded at one second intervals. GPS co-
gondola ordinates allow the instantaneous location on the
Empty 0.0012 105 0.126 0.219 track to be determined. Simulation programs
gondola calculate the acceleration of a train from the force
Lead 0.0024 160 0.384 0.670 balance based on the track geometry, the tractive
loco effort curve, and the instantaneous resistance. This
Trailing 0.00055 160 0.088 0.154 is integrated to give velocity and distance. In this
case, the instantaneous tractive effort time history
loco
is known and becomes an input to the simulation.
Table 3 Air Drag as per Canadian National By trialling different values for the resistance
coefficients, the aim is to match the recorded
The Canadian work suggests that the resistance instantaneous velocity.
for an empty hopper wagon is nearly 3 times that
for a loaded wagon, due to turbulence inside the Using this methodology to verify Davis coefficients
hopper. Also, resistance for the lead locomotive is relies on accurate data for both track geometry
quite high, but it is generally found that trailing and the mass for the vehicles in the consist. It was
locomotives have lower resistance, similar to that found that in practice there may be variations in all
of wagons. of these. Loadings on coal and ore trains vary by a
few percent from the nominal, and while the exact
Total resistance for the train is calculated by weight may be known if the payload is accurately
weighed, it is sometimes a problem to match the

R
summing the drag for each vehicle. However, Lai
and Barkan [5] report studies indicating that drag locomotive logs to that particular train. Also,
decreases until about the 10th unit, and then published track geometry can be surprisingly

Broc
remains approximately constant per unit. inaccurate. Gradients are subject to small changes
as track is maintained and re-laid, and this can be
Again it should be stated that it is incorrect to significant. This method uses data from multiple
include the air resistance in a lb/ton value for each runs to average out the variations, rather than
vehicle, and then to sum over R’ and then multiply trying to get very accurate data for each run.
by the mass of the train. The aerodynamic drag is

SUPPORTED BY
 Frank Szanto ROLLING RESISTANCE REVISITED
Downer Rail

To perform a run, it is necessary to choose a fairly for the loaded case, i.e. 1120 N for 160 tons.
straight section of track, or one with generous However, at 23 tonnes, it gives 424 N, which is
curve radii, as resistance increases in tight curves. below the measured value.
It is necessary to locate the starting point at a
clearly defined feature such as a passing loop, Using the above for A and A0, the air drag values
where the GPS co-ords from the event recorder were found to be 0.045 for the empty case, and
(EVR) can be matched to a point in the track 0.030 for the loaded case (for 10m2 cross-sectional
geometry. There should be no friction braking area). These are significantly lower than the RP-
through the run, as it not possible to accurately 548 value (.0096) and the Canadian National
calculate the actual brake effort from the EVR data. values (0.077 and 0.219 – see Table 3). This
Dynamic braking, however, can be handled if the would support the theory that drag mostly affects
DB Effort is recorded as a force. The section the front dozen vehicles and decreases along the
should also contain no steep grades requiring very train, so that average resistance decreases for
high adhesion, as high creep values will affect the longer trains. Especially significant is the result
results. that, although the air drag is higher on empty
wagons, it is only by around 50%, while the
Canadian National figures suggest it should be
8. IRON ORE TRAIN higher than the loaded case by a factor of three.
The first case considered is for a long train carrying
ore in the Pilbara. Data was obtained for two
locomotives hauling a train with 244 wagons,
approximately 2500m in length. Wagon mass was
23 tonnes tare, and 160 tonnes gross. The A and
A0 terms largely determine the fit at lower speeds,
and then the air drag term can then be adjusted to
fit higher speeds. The very large variation in axle
load presents an opportunity to check the linear fit
over a wide range, with the aim of finding A and
A0 values which are truly constant, and not
‘adjusted’ for empty and loaded conditions as in
some publications. Note that for low to medium
axle loads, the A (weight) and A0 (axle) Fig 1 Loaded 244 wagon ore train
components are similar in magnitude, which makes
it harder to distinguish their relative contributions.

By trial and error, the A value was varied from the

RP-548 value of 6.5 N/tonne, while keeping A0
fixed at 80 N/axle. Figures 1 and 2. below show
plots of tractive effort and speed verses time, and
show the fit between the simulated speed and the
actual speed (from EVR data). A good fit at low
speeds was found with A = 5 N/t for the loaded
case, and 7.5 N/t for the empty case, as in Table 4
below. Recalculating from these data points to get
constants which work for both the empty and
loaded cases gives A = 4.5 N/tonne (= 0.9 lb/ton ~
65% of the RP-548 value), and A0 = 100 N/axle
(=22 lb/axle, slightly higher than the RP-548 Fig 2 Empty 244 wagon ore train
value).

9. HUNTER VALLEY COAL

R
Car Mass A A0 R
Tonne N/t N/axle (N) The second case study is for coal trains in the
160 5 80 1120 Hunter Valley. Train length is about half of the

Broc
23 7.5 80 492.5 Pilbara case, with wagons which are longer, but
Both 4.5 100 similar in tare weight. There are two lengths of
Table 4 Constants for iron ore train trains – 82 and 96 wagons, with loaded weights of
100 tonnes and 120 tonnes per wagon
While there is no reason to expect the SNCF respectively. Both are hauled by three GT46C-ACe
double-deck EMU formula (5) to be valid at this
axle load, it surprisingly gives the exact resistance

SUPPORTED BY
 Frank Szanto ROLLING RESISTANCE REVISITED
Downer Rail

locomotives. The difference in length might be

expected to show up in the air drag value.

The section from Kooragang Island to Hexham is

relatively flat and straight, and it was expected that
this would give good results. However, the
published track data for this section is very
inconsistent, to the extent that the sign of some
grades is wrong.

Figure 3 below shows the results for an empty train

of 82 wagons running from near Hexham to
Metford, a distance of 14 km. The track is almost
straight, with one large radius curve. Using the
previous values for A and A0, it was found that C= Fig 5 Singleton-Belford loaded, 96 wagons
0.09. The curve fit is not as good as for the iron ore
trains.
Figure 5 shows results for a loaded 96 wagon train
in the section from Singleton to Belford. It should
be noted that for this section, the published
gradient data in Manual TS 0002 TI v2 July 1999,
[6], appears to be very inaccurate, and it was
impossible to correlate the results. After wasting
many hours, I found that the data in M-Train file
TS21304 allowed good correlation. M-Train is a
simulation program developed by the PTC NSW
Mechanical Branch in the 1970s – so old that the
curve radii are in chains. I had no expectation that
this data would still be valid or accurate. However,
at several sections where I have tried to compare
simulation results and actual data, the M-Train
data appears to be more accurate in terms of
Fig 3 82 empty wagons Hexham to Metford
where the gradients begin and end, and even the
magnitude of the slope.
Figure 4 below shows results for an empty 96
The results for the Hunter Valley coal trains
wagon train. The coefficient values are low, but
confirm that the rolling resistance components are
they are somewhat meaningless as the correlation
lower than given by RP-548, but that the air drag
between the speed curves is rather poor. This may
terms are close to RP-548. A summary comparing
be due to inaccuracies in the track data, but serves
the experimentally derived coefficients with RP-548
as an example of poor correlation – the method
follows in Table 5.
does not always give reliable results.
A A0 C
N/t N/axle
RP-548 6.5 80 0.096
Ore E/L 4.5 100 0.030/0.045
Coal 4.0 100 0.09
Table 5 Comparison of Constants

10. CONCLUSION

R
Based on the limited number of cases presented
for coal and ore trains, the results consistently
show that the constant for the weight dependent

Broc
term in the Davis equation is around 4 to 4.5
N/tonne, which is lower than in RP-548 (6.5 N/t =
Fig 4 96 empty wagons Maitland to Greta 1.3 lb/t). However, the axle term which is more
dominant for empty wagons is 90-100 N/axle, 10-
20% higher than the 80 N/axle (18 lb/axle) in RP-
548.

SUPPORTED BY
 Frank Szanto ROLLING RESISTANCE REVISITED
Downer Rail

Having said that, these are not the “right” values to comparing the effects of train length. The intent is
use for all cases. Many factors influence rolling to look at intermodal trains too, but this task will be
resistance, including track stiffness, wheel much more difficult due to the varying weights,
diameter, temperature of grease and lozenging of lengths and cross-sectional areas of vehicles
bogie frames. The cases presented are for track making up a general freight train.
and rolling stock in good condition, typical of heavy
11. REFERENCES
haul operations in Australia. Resistance may well
be higher in more general conditions. 1. Davis, W.J jr, 1926. The Tractive
Resistance of Electric Locomotives and
While by no means proving that the velocity term in
Cars, General Electric Review, vol. 29.
the Davis equation is zero, my results support the
contention that it is negligible. 2. Association of American Railroads, AAR
Manual of Standards and Recommended
Results for the aerodynamic drag term are
Practices RP-548, 1960, 1994, 2001.
somewhat more inconclusive. For the limited
number of coal trains examined, the drag seems to 3. American Railway Engineering and
be close to the RP-548 figure, but much lower for Maintenance-of-Way Association, 1999,
long ore trains. It would be necessary to check AREMA Manual for Railway Engineering.
many more cases to reach a conclusion. pp 16-2-4 to 16-2-9.
Part of the problem is that the apparent 4. Kim, SW, Kwon HB, Kim YG and Park TW,
aerodynamic drag is affected by wind speed, so it 2006, Calculation of resistance to motion
is necessary to use data from many runs on of a high-speed train using acceleration
different days to account for this effect. measurements in irregular coasting
conditions. Proc. IMechE Vol. 220, Part
As a methodology, it is shown that deriving
F:J. Rail and Rapid Transit.
resistance coefficients from event recorder data is
viable. However, it is fairly labour intensive. I was 5. Lai, Y.C. and and Barkan, C.P.L, 2005.
intending to include many more cases in this Options for Improving the Efficiency of
paper. Picking track sections and the Intermodal Trains. Transport Research
corresponding sections from event recorder data Board.
was all done manually, and I am giving some
6. Rail Access Corporation, 1999, TS 0002
thought to how this process can be automated.
Also, having erroneous data for track geometry is a TI, v2 Infrastructure Engineering Manual,
Curve and Gradient Diagrams, now
major problem.
available on NSW Asset Standards
Future work is aimed at getting better results for Authority Website.
the aerodynamic drag coefficient, and especially in

R
Broc

SUPPORTED BY