Structural Health Monitoring of Railroad Wheelsyongming.faculty.asu.edu/paper/Microsoft Word - Structural Health... · Structural Health Monitoring of Railroad Wheels Using Wheel

Structural Health Monitoring of Railroad Wheels Using Wheel Impact Load Detectors

Brant Stratman, Yongming Liu, Sankaran Mahadevan*

Vanderbilt University, Nashville, TN 37235, USA

Abstract

This paper proposes two quantitative criteria for removing railroad wheels from

service, based on real-time structural health monitoring trends that are developed using

data collected from trains while in-service. The data is collected using Wheel Impact

Load Detectors (WILDs). These impact load trends are able to distinguish wheels with a

high probability of failure from high impact wheels with a low probability of failure. The

trends indicate the critical wheels that actually need to be removed, while at the same

time allowing wheels that aren’t critical to remain in service. As a result, the safety of

the railroad will be much improved by being able to identify and remove wheels that

have high likelihood of causing catastrophic failures.

1 Introduction

Traditional inspection techniques used in the railroad industry, such as drive-by

inspections where all of the wheels on the train are glanced at while an inspection vehicle

drives by, are not as accurate and reliable as more rigorous and quantitative inspection

methods. Many damaged wheels aren’t found, while many useable wheels are removed

when they could remain in service. By using Wheel Impact Load Detectors (WILDs),

structural health monitoring trends can be developed based on the wheel impact data

which indicates the actual condition of the wheels. The trends can indicate the critical

* Corresponding author, Tel.: 615-322-3040; Fax: 615-322-3365; Email:[email protected]

2

wheels that actually need to be removed, while at the same time allowing wheels that

aren’t critical to remain in service.

The WILD system is composed of a series of strain gages welded to the rail. The

strain gages quantify the force applied to the rail through a mathematical relationship

between the applied load and the deflection of the foot of the rail. These impact forces

are used to monitor locomotive and rail car wheel health to ensure safe train operations.

The use of WILDs by railroads has resulted in the removal of high impact wheels

that damage bearings, lading, rail, other mechanical components, and the wheels

themselves. Although wheels with extremely high impacts are rare on a percentage basis,

small percentages do not equate to small effects. For example, high impact wheels have

been observed to increase the surface crack growth rate on a rail by a factor of nearly 100

times than that under non-impact loading conditions1. Also, it has been shown in a

numerical study that dynamic impacts have a detrimental effect on concrete sleeper

health by increasing the risk of crack initiation2.

High impact wheels are wheels that have an impact force of 90,000 pounds or

greater. High impact wheels are most often wheels that have a flat spot on the tread

surface, known as a slid flat. Slid flats are generally caused when a hand break is left

engaged while the train is in motion; because the wheel cannot rotate, the friction

between the wheel and rail causes the surface of the wheel to become flat instead of

round. Other high impact wheels are caused by major defects in the tread surface; these

high impact wheels have a higher probability of leading to catastrophic failure than slid

flat wheels. It is assumed that derailment is a worst case scenario, but that financial loss

begins well before the vehicle derails due to the high impacts between the wheel and rail.

3

It has been observed that some high impact wheels can remain in service for a

couple of years before failure, while others fail almost immediately3. This indicates that

the high impact force alone is not responsible for wheel failures. Only high impact

wheels that have a defect present pose a significant risk of failure. It has been reported

that the crack growth rate of metallic materials under impact fatigue is significantly

higher than under non-impact fatigue4,5.

Currently the Association of American Railroads’ (AAR) wheel impact load limit

for wayside detectors is 90,000 pounds6. Early experience with wheel impact load

detectors showed that a large percentage of high impact wheels were not condemnable

under AAR rules7, although such detectors were effective at finding high impact wheels.

Many of the most catastrophic wheel failure modes, namely shattered rim failures

and vertical split rim failures, occur below the 90,000 pound condemning limit making

the WILD system ineffective when looking at individual impacts.

This paper proposes two additional criteria for removal of wheels with high

likelihood of failure, based on two real-time structural health monitoring trends that were

developed using data collected from in-service trains. These impact trends are able to

distinguish wheels with a high probability of failure from high impact wheels with a low

probability of failure. As a result, the safety of the railroad will be much improved by

being able to identify and remove wheels that have the highest likelihood of causing

catastrophic failures.

2 Wheel Impact Load Detectors A Wheel Impact Load Detector (WILD) is used to manage the wheel impact load

spectrum for targeted removals of defective wheels from service. The WILD continually

4

monitors locomotive and rail car wheel health to ensure safe train operations. WILDs

scan millions of wheels per day throughout the international rail industry.

The installation of WILDs requires no radical modification of the existing track

structure. A series of strain gage load circuits, micro-welded directly to the neutral axis

of a rail, create an instrumented zone for the measurement of vertical forces exerted by

each wheel of a passing train. Signal processors, housed in a nearby unit, electronically

analyze the data to isolate wheel tread irregularities (see Figure 1). If any wheel

generates a force that exceeds a tailored alarming threshold, a report identifies that wheel

for action. A low-level alarm identifies cars for service at the next available opportunity;

a mid-level alarm limits a train's maximum speed until the car can be set aside for the

wheel to be removed; and a high-level alarm directs a train to stop as quickly and safely

as possible to avoid a potential derailment. These reports are usually distributed in real-

time to such interested parties as rail traffic control centers and car shops.

There are two widely used noncontact methods of quantifying the extent of

damage on the wheel8: strain-based systems, and accelerometer based systems. The

strain-based system quantifies the force applied to the rail through a mathematical

relationship between the applied load and the deflection of the foot of the rail. Strain in

the rail foot is used as a direct measure of the load at the rail head. The accelerometer

based system uses accelerometers placed on the rail to infer the extent of damage on the

wheels from the accelerometer’s output. In some cases, the accelerometer output is

converted to a force. The strain-based system is used for load measurements in this

study.

5

Figure 1 Schematic of a WILD site.

One setup for a strain gauge-based wheel impact monitor consists of 11 strain

gauge cribs, per rail9,10,11. The gauges are bonded beneath the foot of the rail and the

applied force was determined from the shear strain. Some overviews of a typical strain

gauge wheel detection system can be found in1,10,11. Fundamentally, the system consists

of strain gauges instrumented as shown in Figure 2, on both sides of the rail web with the

middle of the strain bridge being in the center of a crib.

Figure 2 Rail web schematic; full-bridge setup.

300mm

Front Side

Neutral axis

1 2

3 4

6

The stain gauges are then wired into a full-bridge circuit where both sides of the web are

taken into account to nullify the effects of any off-center strikes. According to the

arrangement as shown in Figure 2, the output of the bridge is then defined by the

following equation:

( ) ( ) ( ) ( ) ( )[ ] CFVttttGF

tFout ⋅

−+−= 4321

4εεεε (1)

where Fout(t) is the output force and εn(t) are the respective dynamic output strains; both

are functions of time. GF is the gauge factor as specified by the manufacturer of the

strain gauges, V is the input excitation voltage, and CF is a calibration factor.

Traditionally, CF is simply obtained by quasi-statically loading the instrumented

structure prior to testing.

For the work presented in this paper each WILD site was made up of 128 strain

gages; 32 strain gages for vertical force measurements and 32 strain gages for lateral

forces measurements, per rail. In this setup two strain gages are required to give one

vertical force measurement and two strain gages are required to give one lateral force

measurement (Figure 3).

Each WILD site collected 16 vertical force values and 16 lateral force values per

rail (according to the instrumented zones in Figure 1); the strain gages were spaced for

maximum exposure to all wheel sizes. For most wheel sizes the force between the wheel

and rail was collected for 90% of the wheel revolution.

In order to reduce data storage space the 16 vertical and 16 lateral force values for

each rail were used to calculate a vertical average force, vertical peak force, lateral

average force, and lateral peak force measurement for each wheel. This paper focuses

7

solely on the vertical average and vertical peak forces; these loads will be referred to as

the nominal (average) and peak loads throughout the rest of the paper.

Figure 3 Strain gage setup for this study.

The nominal load is the average of the 16 WILD impact measurements for the

wheel. The peak load equals the peak reading of the 16 WILD impact measurements. A

high peak load compared to nominal load would indicate a non-round (defect) spot on the

wheel, while a peak load equal to the nominal load would indicate a perfectly round

wheel.

The nominal loads of all eight wheels on a train car can be used to identify

unbalanced/shifted loads and overloads. For a full 286K GRL freight car, the load per

wheel is 35,750 pounds. Therefore, if a wheel impact load detector recorded an impact

reading of 90,000 pounds, the applied dynamic load is 54,250 pounds in addition to the

normal (nominal) wheel load12. The ratio of the peak load to the nominal load can be

used to detect wheel defects in an empty car13. Two semi-normalized impact forces, the

dynamic impact load and ratio, are calculated from the following formulas:

8

( ) ( )load nominalloadpeak loadimpact Dynamic −= (2)

( )

( )load nominal

loadpeak Ratio = (3)

The dynamic impact load and ratio are referred to as semi-normalized impact

forces because they cancel out the effect the weight has on the impact reading. For

example, a wheel carrying more weight will have a higher peak load simply because

there is more force on the wheel. Similarly, a train moving at a higher speed will have a

higher peak load simply because there is more force on the wheel. Since the dynamic

impact load and ratio cancel out the effect the weight of the car has on impact reading,

but the speed of the train still affects their readings, these are considered only semi-

normalized impact values.

3 Wheel Failures and Failure Modes

In recent years, higher train speeds and increased axle loads have led to larger

wheel/rail contact forces. Also, efforts have been made to optimize wheel and rail

design. This evolution tends to change the major wheel rim damage from wear to

fatigue14. Unlike the slow deterioration process of wear, fatigue causes abrupt fractures

in wheels or tread surface material loss.

Railroad wheels may fail in different ways corresponding to different failure

mechanisms15,16,17. Shattered rims and vertical split rims account for the majority of

wheel failures in North America.

Shattered rim defects are the result of large fatigue cracks that propagate roughly

parallel to the wheel tread surface19,18. They form and grow ½ to ¾ inch (12-20 mm)

9

below the tread surface19. They can grow up to a length of several hundred millimeters.

Ekberg et al reported that a shattered rim can initiate from both inclusions and non-

inclusion locations20. An examination of shattered rim fracture surfaces normally reveals

a clamshell pattern as seen in Figure 4. If service damage is not too extensive, the crack

initiation site is often clearly visible.

Figure 4 Typical Shattered Rim; arrow points to origin.

Once shattered rims initiate in the wheel rim, propagation occurs rapidly under

normal wheel loading. Studies have shown that to initiate cracking a large load, such as

an impact, may be required21,22,23. Shattered rims tend to occur in either relatively new

wheels or those close to their condemning limit.

Vertical split rims are another type of wheel failure that is believed to be related

to high impact loads. A vertical split rim is believed to be due to strength failure; the

failure is a sudden failure characterized by vertical crack propagation where the tread

surface of the wheel is detached.

A large amount of scatter has been observed in the fatigue life distribution of

railroad wheels, ranging from several months to several decades (Figure 5).

10

0

0.2

0.4

0.6

0.8

1

5 6 7 8 9 10

Fatigue life (log(N))

Pro

bab

ility

Field data

0

0.05

0.1

0.15

0.2

0.25

0.3

5 6 7 8 9 10

Fatigue life (log(N))

Fre

qu

en

cy

Field data

Figure 5 Empirical CDF and frequency histogram of wheel field data.

In Figure 5, it is seen that the field data has two ranges of fatigue life distribution.

This phenomenon can be clearly seen in the frequency diagram (Figure 5(b)). It also can

be seen that the number of wheels experiencing premature failure is only a small fraction

of wheels (around 10%) and does not greatly affect the overall mean fatigue life.

However, their effects are significant at the tail region, which affects the reliability

evaluation.

Berge24 and Stone and Geoffrey25 suggest that the large stress, perhaps due to

wheel/rail impact or material discontinuity, has important effect on the shattered rim

failure. Also, the large on-tread brake loading and the thermal stresses arising from on-

tread friction braking will reduce the fatigue life of railroad wheels26. The observed

premature fatigue failure (earlier failure modal in Figure 5(b)) is possibly due to the

above mentioned factors or their combinations, such as initial defects, brake loading, and

thermal loading.

From the discussion of the failure modes and the fatigue failure life data, and also

from Figure 5, it is clear that catastrophic wheel failures can occur suddenly at any point

during the wheels’ lifetime. For this reason, the railroad industry will benefit from real-

a) CDF b) Frequency histogram

11

time structural health monitoring trends that are based on wheel impact data and monitor

wheels throughout their entire lifetime. The following sections of the paper develop two

such trends.

4 Comparison between High Impact Wheels and Failed Wheels

Currently, wheel impact load detectors are used only to identify wheels with

impacts greater than 90,000 pounds. A review of the lab database of a major North

American railroad revealed that impacts for wheels were beginning to increase over time,

even though the forces were still far below the action level of 90,000 pounds. The rises

in wheel impacts were caused by the progression of a crack which created an out-of-

round condition to the wheel3. This observation prompted research to find some

predictive characteristics, distinguishable from high impact wheels, which would identify

a wheel in real-time with a high probability of failing.

An initial investigation of the wheel impact data found that failed wheels and their

mate wheels had significantly different impact readings. Thus, one possible approach is

to remove wheels with significantly different impact readings from their mate wheels.

However, in the case of high impact wheels, since they are generally wheels with a slid

flat, both wheels on the axle have high impacts. Further analysis of the data determined

that removing wheels based on difference from mate wheel impact is not a cost efficient

method for identifying wheels with a high probability of failure because this is not a

unique enough characteristic among the wheel population; based on the analysis,

approximately 2400 wheelsets per day, in North America, have a significant impact

difference between the two wheels on the wheelset.

12

A more rigorous analysis is performed in this paper comparing high impact

wheels and failed wheels and has identified distinguishing characteristics between them.

The result of the statistical summary comparing high impact wheels with failed wheels is

shown in Table 1.

Table 1 Failed wheels get worse significantly faster than high impact wheels.

Av

era

ge

Ma

xim

um

Dy

na

mic

(k

ips)

Av

era

ge

Ma

xim

um

Ra

tio

Av

era

ge

dy

na

mic

(in

clu

des

lo

ad

ed

an

d e

mp

ty)

bef

ore

loa

d i

ncr

ease

(k

ips)

Wee

ks

ab

ov

e a

dy

na

mic

im

pa

ct o

f

20

kip

s

Len

gth

of

tim

e fr

om

av

era

ge

dy

na

mic

to

pea

k i

mp

act

(w

eek

s)

High Impact Wheels 63.31 6.44 13.40 41.00 30.05

Failed Wheels 44.53 4.02 13.35 14.75 4.15

From Table 1 it is seen that high impact wheels are actually worse than broken

wheels in every category except length of time to peak impact (the shaded cells indicate

which wheels are worse in each category). It is seen that high impact wheels have higher

impact readings for a longer period of time than broken wheels, however broken wheels

get worse much faster. This sudden increase indicates rapid crack growth leading to

failure.

It is also important to note that some trains run 100 miles/week while others run

2,000 miles/week. Therefore an analysis was conducted in order to observe whether

wheels that failed were on more active trains. This analysis determined whether failed

wheels’ impacts actually increased more rapidly (fewer revolutions) than high impact

wheels, or if they simply were on more activate trains giving the illusion that they

13

increase more rapidly. The result of the statistical summary comparing high impact

wheels with failed wheels is given in Table 2.

Table 2 Mileage comparison between failed wheels and high impact wheels.

Mil

eag

e f

rom

wh

ee

l m

ou

nt

da

te u

nti

l d

ev

iati

on

fro

m

av

era

ge i

mp

ac

t (m

ile

s)

Mil

eag

e f

rom

wh

ee

l tu

rne

d

da

te u

nti

l d

ev

iati

on

fro

m

av

era

ge i

mp

ac

t (m

ile

s)

Mil

eag

e f

rom

av

era

ge

d

yn

am

ic t

o p

ea

k im

pac

t (m

ile

s)

To

tal

mile

ag

e f

rom

wh

ee

l m

ou

nt

da

te u

nti

l p

ea

k

imp

ac

t (m

iles

)

To

tal

mile

ag

e f

rom

wh

ee

l tu

rne

d d

ate

un

til p

eak

im

pa

ct

(mil

es

)

Av

era

ge

mil

es

pe

r ye

ar

(mil

es

)

High Impact Wheels 356046.0 89911.1 26868.42 395755.1 124000.0 45713.7 Failed Wheels 341814.4 146050.0 3900.00 346604.5 151525.1 41590.4

From Table 2 it is seen that the mileage analysis agrees with the time analysis;

broken wheels’ impacts increase much more rapidly than high impact wheels. It is also

important to observe in Table 2 that high impact wheels and broken wheels, on average,

travel relatively the same distance per year. It is therefore concluded in this study that

since high impact wheels and broken wheels travel the same distance in the same amount

of time on average, then using time intervals is appropriate for developing the structural

health monitoring trends.

5 Structural Health Monitoring Trends Using Wheel Impact Load Detectors

Two trends are developed that could be used to monitor wheels in real-time. The

two trends identify wheels that suddenly get worse, thus providing a way to differentiate

wheels with a high probability of failure from high impact wheels. In general, high

impact wheels appear to degrade in a gradual manner, as seen in Figure 6.

14

From Figure 6 it is seen that wheels generally have an up-down-up impact trend,

which is caused by the train’s loaded/empty cycle for moving cargo. This means that

when a train is carrying cargo the impact is higher than when the train is empty, hence the

up-down-up impact trend. When a defect in a wheel starts to grow, the authors found that

the impacts will increase regardless of whether or not a train is carrying cargo. These

defect growth trends can be seen in Figures 7 and 8, when consecutive impacts increase

rather than following the up-down-up trend.

Figure 6 High impact wheels generally increase impact readings slowly over time.

5.1 First Structural Health Monitoring Trend

The first trend developed is designed to capture wheels that were running at a

high dynamic impact but increased greatly over a short period of time. This trend

differentiates wheels with a high probability of failure from high impact wheels since

0

10

20

30

40

50

60

70

4/2

0/2

003

6/9

/2003

7/2

9/2

003

9/1

7/2

003

11/6

/2003

12/2

6/2

003

2/1

4/2

004

4/4

/2004

5/2

4/2

004

Time

Dyn

am

ic (

kip

s)

High Impact Wheel Dynamic

Mate Wheel Dynamic

15

high impact wheels’ impacts increase slowly over time. Generally all of the wheels

caught by the first trend have peak impacts greater than 90,000 pounds so they are

already condemnable by AAR standards.

The first wheel removal criterion is found to be as follows, based on analysis of

the data. The wheel needs to have three consecutive increasing impact readings, each

reading with an impact increase of 15 kips or more than the previous reading, and all

three readings need to occur in a period of less than 50 days with the peak reading having

a dynamic impact greater than 70 kips. An example of the first trend is given in Figure 7.

Of the failed wheels included in the database, 10.53% (14.28% of Vertical Split

Rims, and 8.33% of Shattered Rims) had impact histories following the trend discussed

above. This means that 10.53% of wheel failures would have been prevented by the first

structural health monitoring criterion had it been implemented at the time of the wheels’

service.

It is estimated from the data that approximately 200 wheels per year in North

America are captured by this wheel removal criterion. Examining data from previous

years concludes that of the wheels that would have be caught by the first criterion had it

been implemented, 78.2% were removed for another reason (i.e. high impact wheel as

detected by wheel impact detector, visual inspections, etc.) after the criterion would have

detected them. This means that because the criterion was not in place 156 damaged

wheels were left in service and caused additional damage to the rail and train

components. According to the data, the damaged wheels would have been removed 31.7

days earlier on average had the first criterion been implemented at the time of their

16

operation, which in effect would prevent 31.7 days of additional damage from those

individual wheels.

Implementing this criterion would increase the number of removed wheels by

approximately only 44 wheels/year; this equals the 200 wheels/year that are detected by

the criterion minus the 156 wheels/year that would be removed regardless of whether the

criterion was in place or not.

Figure 7 An example of the first trend showing rapid increase in impact readings.

The number of shattered rims has been estimated to be around 700 wheels per

year in North America12; implementing the proposed criterion would prevent 8.33% of

those. This means approximately 58 shattered rim failures per year will be prevented by

increasing the number of wheels removed per year by only 44 wheels. As discussed

above, actually the trend detects approximately 200 wheels/year, approximately 58 of

0

10

20

30

40

50

60

70

80

90

100

110

3/1

2/2

00

3

4/1

2/2

00

3

5/1

2/2

00

3

6/1

2/2

00

3

7/1

2/2

00

3

8/1

2/2

00

3

9/1

2/2

00

3

10

/12

/20

03

11

/12

/20

03

12

/12

/20

03

1/1

2/2

00

4

2/1

2/2

00

4

3/1

2/2

00

4

4/1

2/2

00

4

5/1

2/2

00

4

Time

Dy

na

mic

(k

ips

)

Failed Wheel Dynamic

Mate Wheel Dynamic

17

which are shattered rims, but because some of the shattered rims wouldn’t have been

detected until after they failed the proposed criterion actually prevents more shattered

rims than additional wheels it would require to be removed. It is important to note that

included in the 200 wheels detected by the trend each year are a significant number of

vertical split rim failures that will also be prevented.

The average cost of a wheelset (including new and refurbished wheelsets) is

approximately $830. Therefore, the additional 44 wheelsets that would be removed by

this tread would cost North American railroads approximately $36,520. Considering the

damage these failed wheels cause to the rail and train components, and their high

probability of causing derailments, implementing this trend would be financially

beneficial and at the same time increase the safety of the railroad.

5.2 Second Structural Health Monitoring Trend The second trend is developed to capture wheels that are running in a normal

impact reading range, and suddenly increase in a very short period of time. Wheels

following the second trend have a dynamic impact history that looks nothing like the

dynamic impact history of a high impact wheel.

The second criterion for wheel removal is found to be as follows, based on

analysis of the data. The wheel needs to have three consecutive increasing impact

readings, each reading with an impact increase of 2 kips or more than the previous

reading with an average increase of 10 kips or more for the three readings, and all three

readings need to occur in a period of less than or equal to 20 days with the peak reading

have a dynamic impact greater than 40 kips, no reading 30 days prior the initial impact

18

deviation reading can have an impact reading greater than 5 kips or less than 15 kips from

the initial deviation dynamic value. An example of the second trend is given in Figure 8.

Of the failed wheels included in the database, 5.26% (8.33% of Shattered Rims)

had impact histories following the trend discussed above. This means that 5.26% of

wheel failures would have been prevented by the second structural health monitoring

criterion had it been implemented at the time of the wheels’ service.

Figure 8 An example of the second trend showing sudden increase in impact readings.

It is estimated from the data that approximately 1400 wheels per year in North

America are captured by this wheel removal criterion. Examining data from previous

years concludes that of the wheels that would have be caught by the second criterion had

it been implemented, 47.5% were removed for another reason after the criterion would

have detected them. This means that because the criterion was not in place 665 damaged

0

10

20

30

40

50

3/1

5/2

003

4/1

5/2

003

5/1

5/2

003

6/1

5/2

003

7/1

5/2

003

8/1

5/2

003

9/1

5/2

003

10/1

5/2

003

11/1

5/2

003

12/1

5/2

003

Time

Dynam

ic (kip

s)

Failed Wheel Dynamic

Mate Wheel Dynamic

19

wheels were left in service and caused additional damage to the rail and train

components. According to the data, the damaged wheels would have been removed 66.6

days earlier on average had the second criterion been implemented at the time of their

operation, which in effect would prevent 66.6 days of additional damage from those

individual wheels. Implementing this criterion would increase the number of removed

wheels by approximately 735 wheels/year, thereby preventing damage from these wheels

to the train and rail components.

The average cost of a wheelset (including new and refurbished wheelsets) is

approximately $830, and the average cost of a derailment is approximately $340,000.

This means only two derailment preventions per year from the 1400 wheels following

this trend would cover the cost of the additional 735 removed wheelsets detected by this

criterion, and at the same time would increase the safety of the railroad. In addition,

when the additional damage to the rail and train that is prevented from this criterion is

included in the cost analysis it is clear that this structural health monitoring trend

provides a cost efficient criterion for removing wheels.

From the above discussions it is clear that both criteria are cost efficient methods

for removing wheels. Even so, when implementing the two criteria together the cost

efficiency is further increased; a very small percentage (0.57%) of the wheelsets are

identified by both criteria. The combined implementation of the two criteria will result in

approximately 775 removed wheelsets per year. This slight overlap in wheel detection

reduces the increase in removals by 4 wheelsets/year; making the two trends, when

implemented together, an even more cost efficient method for removing wheels.

20

6 Conclusion

Traditionally, wheels are removed based on drive-by visual inspection and the

90,000 pound high impact condemning limit. Unfortunately many damaged wheels are

not found with the drive-by inspection method, while many useable wheels are removed

when they could remain in service. The structural health monitoring trends developed in

this paper provide a quantitative decision method based on the wheel impact data which

indicates the actual condition of the wheels. Therefore the trends indicate the critical

wheels that actually need to be removed, while at the same time allowing wheels that

aren’t in critical condition to remain in service.

The two wheel removal criteria proposed in this paper will remove 15.8%

(16.67% of Shattered Rims, and 14.28% of Vertical Split Rims) of wheels with a high

probability of failing by indicating that the wheels are about to fail before they actually

fail. The two criteria will increase the number of wheels removed, for the accumulation

of railroads in North America, by approximately 775 wheelsets/year. Considering the

damage these failed wheels cause to the rail and train components, and their high

probability of causing derailments, implementing these criteria would be financially

beneficial while at the same time increase the safety of the railroad.

7 Acknowledgments

The research reported in this paper was supported by funds from Union Pacific

Railroad and Meridian Railroad (Research Agreement No. 18140, Monitors: Rex Beck

and Todd Snyder). The support is gratefully acknowledged.

21

References 1 Lee, M.L. and Chiu W.K. (2005). A comparative study on impact force prediction on a railway track-like structure. Structural Health Monitoring, Vol. 4, No. 4, 355-376. 2 Gustavson, R. (2000). Static and dynamic finite element analysis of concrete sleepers. Licentiate of Engineering Thesis, Chalmers University of Technology, Division of Concrete Structures, Göteborg. 3 Union Pacific’s Lab Database. 4 Tanaka, T., Kinoshita, K. and Nakayama, H. (1992). Effect of loading time on high cycle range impact fatigue strength and impact fatigue crack growth. JSME International

Journal, Series 1, 35(1), 108-116. 5 Cheng, Y., Chen, D. and Nogata, F. (1994). Fatigue behaviour of a rail steel under low and high loading rates. Fatigue and Fracture of Engineering Materials and Structures, 17(1), 113-118. 6 Association of American Railroads, 2005 Field Manual of the AAR Interchange Rules, Rule 41, AAR, Washington, DC. 7 S. Kalay and A. Tajaddini, “Condemning Wheels Due to Impact Loads: Preliminary Survey – Six Railroads Experience,” AAR Report R-754, February, 1990. 8 Barke D, Chiu WK. Structural Health Monitoring in the Railway Industry: A Review. Structural Health Monitoring 2005; 4(1):81-93. 9 Johansson, A. and Nielsen, J. (2001). Railway wheel out-of-roundness – Influence of wheel-rail contact forces and track response. 13

th International Wheelset Conference,

Rome. 10 Kalay, S., Tajaddini, A. and Stone, D.H. (1992). Detecting wheel tread surface anomalies. Rail Transportation – 1992 American Society of Mechanical Engineers, Rail Transportation Division (Publication) RTD, ASME, New York, NY, USA: 165-174. 11 Partington, W. (1993). Wheel impact load monitoring. Proceedings of the Institution of

Civil Engineers – Transport, 100(4): 243-245. 12 Lonsdale, C., Pilch, J. and Dedmon S. (2002). Stress effects of wheel impact loads. –unknown-. 13 Bladon, K. (2003). Barke, D.W. (ed.), Function of Wayside Wheel Impact Monitors, Adelaide, Australia. 14 Tournay, H.M. and Mulder, J.M., “The transition from the wear to the stress regime”, Wear, Vol.191, pp.107-112, 1996. 15 Stone, D.H. and Moyar G.J., “Wheel shelling and spalling – an interpretive review”, in Rail Transportation 1989, ASME, pp.19-31, 1989. 16 Marais, J.J, “Wheel failures on heavy haul freight wheels due to subsurface effects”, Proc 12

th International Wheelset Congress, Qingdao, China, pp. 306-314, 1998.

17 Mutton, P.J., Epp, C.J. and Dudek, J., “Rolling contact fatigue in railway wheels under high axle loads”, Wear, Vol. 144, pp. 139-152, 1991. 18 Giammarise, A. W., and Gilmore, R. S. “Wheel Quality: A North American Locomotive Builder’s Perspective”, GE Research & Development Center, CRD140, Sep, 2001. 19 Stone, D.H., Majumder, G. and Bowaj, V.S. (2002). Shattered rim wheel defects and the effect of lateral loads and brake heating on their growth. ASME International

22

Mechanical Engineering Congress & Exposition, Nov 17-22 2002, New Orleans, Louisiana: 1-4. 20 Ekberg, A., E. Kabo, and H. Andersson., “An engineering model for prediction of rolling contact fatigue of railway wheels”, Fatigue & Fracture of Engineering Materials

and Structures, Vol. 25, pp. 899-909, 2002. 21 Stone, D.H. (2000). Wheel shattered rims – and interpretive review. Wheels and Axles,

Cost Effective Engineering. ImechE Seminar Publication: 75-84. 22 Beretta, S., Roberti, A., Ghidini, A. and Donzella G. Deep shelling in railway wheels. Proceedings, 13th International Wheelset Congress, Rome, Sept 2001, Paper 23. 23 Stone, D.H., Kalay, S.F., and Lonsdale, C.P. Effect of wheel impact loading on shattered rims. Proceedings, 13th International Wheelset Congress, Rome, Sept 2001, Paper 2. 24 Berge, S., “Shattered Rim Fracture Research”, Proceedings of 2000 Brenco Rail Conference, LaQuinta, California, October 19-20, 2000. 25 Stone, D.H., and Geoffrey, E.D., “The Effect of Discontinuity Size on the Initiation of Shattered Rim Defects,” ASME Transportation Division- Vol. 19, pp. 7-14, ASME Spring 2000. 26 Gordon, J., and Perlman, A. B., “Estimation of Residual Stresses in Railroad Commuter Car Wheels Following Manufacture”, Proceedings, International Mechanical

Engineering Congress and Exhibition, ASME RTD Vol. 15, pp.13 – 18, 1998.