Anda di halaman 1dari 7

See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/267600494

A Novel and Reliable Track Condition Prediction


Model for Condition Based Track Maintenance

Conference Paper · April 2013


DOI: 10.1115/JRC2013-2411

CITATIONS READS

0 41

5 authors, including:

Peng Xu Reginald R. Souleyrette


Beijing Jiaotong University University of Kentucky
19 PUBLICATIONS 93 CITATIONS 89 PUBLICATIONS 530 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

US Road Assessment Program View project

Quantifying Rail-Highway Grade Crossing Roughness View project

All content following this page was uploaded by Reginald R. Souleyrette on 12 April 2016.

The user has requested enhancement of the downloaded file.


Proceedings of the 2013 Joint Rail Conference
JRC2013
April 15-18, 2013, Knoxville, Tennessee, USA

JRC2013-2411

A NOVEL AND RELIABLE TRACK CONDITION PREDICTION MODEL


FOR CONDITION BASED TRACK MAINTENANCE

Peng Xu*,1,2 Rengkui Liu


* Corresponding Author: xu.peng.bjtu@gmail.com MOE Key Laboratory for Urban Transportation Complex
1 MOE Key Laboratory for Urban Transportation Complex Systems Theory, Beijing Jiaotong University
Systems Theory, Beijing Jiaotong University Beijing, China
Beijing, China
2 Department of Civil Engineering, College of Engineering,
University of Kentucky
Lexington, KY, United States of America

Quanxin Sun Reginald R. Souleyrette Jerry G. Rose


MOE Key Laboratory for Urban Department of Civil Engineering, Department of Civil Engineering,
Transportation Complex Systems College of Engineering, University of College of Engineering, University of
Theory, Beijing Jiaotong University Kentucky Kentucky
Beijing, China Lexington, KY, United States of Lexington, KY, United States of
America America

ABSTRACT The key to effective implementation of CBM is reliable


forecasts of future conditions based on prediction models. In
Recent railway transportation developments throughout the
this paper, a novel track condition prediction model is
world have demonstrated two main trends, high speed and
presented which may serve as a basis for condition based track
heavy haul. Both of these have resulted in increased wheel
maintenance. The model is built on practical knowledge of
loads due to increased dynamic forces and/or higher weights. It
track condition deterioration. Typically, the model can predict
is well known that increased wheel loads result in faster
track condition (including isolated geometry exceptions and
deterioration of the track structure. Consequently, maintenance-
condition of unit track sections) two to three months in
of-way departments inspect more frequently to ensure safety
advance, depending on tonnage/frequency of trains. To validate
and comfort for passengers and reduce the risk of damage to
the model, track inspection data were collected from the Jinan
freight. An alternative to more frequent inspections is a track
bureau of China Railroads. Some analysis of the results of track
maintenance strategy known as condition based maintenance
condition predictions is also presented.
(CBM). CBM has received considerable attention in other
industries such as truck fleet management and power systems
Keywords: Railroad; Track Condition; Prediction Model;
facility management. Practices in these fields show that CBM
Condition Based Maintenance
can not only reduce interruption of service but also enhance
system reliability. What is more, CBM can also reduce life-
cycle costs. Within railroading, CBM is used to schedule INTRODUCTION
preventive rail grinding, but, CBM has not yet found
widespread implementation in the maintenance of other track To meet the demands of society and economy development
for safe, efficient and economic transportation, railroad industry
components.
employs new technologies to expand its capacity. There are

1 Copyright @ 2013 by ASME


three main ways to increase rail capacity: raising travel speed, in most countries is equal to 1435mm, but on a curved track its
increasing weights and increasing traffic density. For instance, theoretical value is equal to the sum of the designed gauge and
the recorded maximum operational speed for passenger rail is the designed gauge widening. On a straight track the cross
350km/h which is reached on the Beijing-Shanghai High Speed level’s theoretical value is 0mm, but on a curved track it is
Rail line. Because the traffic density cannot be increased equal to the designed super-elevation. Due to some reasons [5, 6]
constantly and thus is limited, feasibly expanding railroad the actual values of these parameters deviate from their
capacity must be done by raising rail travel speeds and/or theoretical ones. These deviations are usually referred to as
carrying more freight. It follows that recent rail transportation track irregularity. Therefore, track irregularity is usually
developments have been characterized by high-speed and indicative of track condition.
heavy-haul trends. Track irregularity under wheel loads is usually measured
It is reported by Selig in reference [1] that dynamic forces with track geometry car [2, 3]. In China railroads, there are three
resulting from increased speed might be three times as large as kinds of track geometry cars [7]. GJ-4 is an extensively used
those caused by the same weight at lower speed. As rail travel kind. Its sampling interval is 0.25cm. It is suitable for
speed and carrying weights increase, wheel loads on the rail inspecting railroads with allowable speeds below its design
contact surface grow considerably due to increased dynamic speed. Besides track geometrical parameters, some additional
forces and/or higher weights. And, the larger the wheel loads on parameters are also measured at the same time, as listed in
a track structure, the faster the track structure deteriorates. In Table 1.
current practice, to ensure track condition remains at acceptable In China, two main methods are employed to quantify
levels for efficient and safe transportation, more frequent track track condition according to measurements of track geometry
inspections are conducted by maintenance-of-way departments car [2]. One is to evaluate the local track condition via extreme
[2]
. values of isolated geometry exceptions. Another is to assess the
An alternative to more frequent track inspections may be to condition of a track section, typically 200m, via the standard
develop a track deterioration model to predict track condition. deviation based method. Usually track condition indicator
To this end, this article presents a track condition prediction obtained through standard deviation is called Track Quality
model which is built on practical knowledge of track condition Index (TQI), as shown in Eq.(1). In the section of error analysis
deterioration. For clarity, background on track condition is on track condition predictions, TQI is used.
firstly given. An introduction to the practical knowledge of  7

track condition deterioration follows. Next, the track condition TQI  i 1


i

prediction model is presented. To illustrate its practicability,  1 n 2 2
errors in track condition predictions are analyzed. Finally,  i   ( xij  xi ) (1)
 n j 1
conclusions are drawn and presented.  1 n
 xi   xij
BACKGROUND ON TRACK CONDITION  n j 1
Wherein:
Track condition often refers to track geometry condition
i = 1,2,3,4,5,6,7 which represent gauge, cross level, left
which is described by several geometrical parameters [3]. These
surface, right surface, left alignment, right alignment,
parameters comprise projections of rails in the horizontal,
and twist,
longitudinal and vertical planes and include gauge and cross
i = the standard deviation of measurements of a given
level in the horizontal plan, surface in the vertical plan,
geometrical parameter i over a given track length,
alignment in the longitudinal plane, and twist for measuring
xi = the average of measurements of a given geometrical
uneven running surface [4]. The theoretical values of gauge and
parameter i over the given track length,
cross level are different between the straight track and the
xij = the measurement value of a given geometrical
curved track, whereas the other three parameters, surface,
parameter i at the j th sampling point within the
alignment and twist, have the equal theoretical values for the
track length,
straight and curved tracks, i.e. 0mm, 0mm, and 0mm/ l m
n = the number of sampling points within the track length.
wherein l is the base length for measuring twist. On a straight
track the theoretical value of gauge is the designed gauge which

2 Copyright @ 2013 by ASME


TABLE 1. MEASUREMENT ITEMS OF GJ-4 TRACK GEOMETRY CAR
No. Item Accuracy Unit No. Item Accuracy Unit
1 Gauge ±0.1 mm 7 Body acceleration ±0.01 g
2 Curvature ±0.05 °/30m 8 Axle acceleration ±1 g
3 Crosslevel 9 Automatic location
±1.5 mm Volt
(Superelevation) detection (ALD)
4 Surface (left, fight) ±1.5 mm 10 Speed ±0.2 km/h
5 Alignment (left, right) ±1.5 mm 11 Milepost Km
6 Twist ±1.5 mm/ l m

PRACTICAL KNOWLEDGE OF TRACK CONDITION


DETERIORATION THE NOVEL TRACK CONDITION PREDICTION
MODEL
Directly influenced by wheel loads, track condition
Based on the above introduced practical knowledge of
deteriorates. But its deterioration rate is determined not only by
the size of the wheel loads but by many other factors also, for track condition deterioration, the track condition prediction
example, track facilities’ age and condition, maintenance model is presented in this section. For model building, some
essential variables are defined firstly.
efficiency, traffic density, and among others. All these factors
influencing track condition deterioration fall into four Variable denotations
categories: track conditions, transportation conditions,
Variable denotations throughout this section are listed as
maintenance conditions, and environmental conditions [6].
follows.
Because influences of these factors spread spatially differently,
ti a track irregularity parameter on which track
the track condition deterioration process at a track position is
condition prediction model is built.
distinctive. And what is more, the evolution process of a track
tn , x the date of the x th day after the nth inspection.
irregularity parameter at a track position is distinguishable.
Therefore, an ideal track condition prediction model should Here, x might be any integer and n be any
be built on the deterioration process of each irregularity nonnegative integer. When x takes on a negative
parameter at a track position. But in reality it is infeasible due integer, tn , x denote the date of the | x |th day before
to the fact that there are mileage errors existing in measurement the nth inspection, wherein | x | is the absolute
data of track geometry car, sometimes even larger than 200m [6, value of x ; when x equals 0, tn ,0 is usually
9]
. Although we have made attempts to develop models to
signified by t n for simplicity and indicates that an
reduce mileage errors as much as possible from two
inspection was conducted on this day.
dimensions, key equipment [6, 9] and historical measurement
anl , x the actual value of ti at the l th sampling point
data [6, 10]. Performance analyses for these models show that
they outperform existing models. However, measurement data within the concerned track length on tn , x . It is worth
processed with our mileage error reduction models still do not noting that only anl ,0 has the measurement value.
meet ideal track condition prediction demands that there are no ^l
mileage errors in measurement data. Consequently, the track a n, x predicted value of anl , x .
condition prediction model should be built for small track  a1n ,0  a1n , p 
lengths, but not for a track position. According to remaining  
An       , wherein p denotes the number
mileage errors in measurement data processed by our models,
 al 
 n ,0  an , p 
l
0.5m is the length of short track sections on which the
prediction model is built. of days on which track condition predictions are
Because track structure over a small track length, such as made.
0.5m, cannot be degraded considerably differently, sampling  ^1 ^1 
 a n ,0  a n, p 
   
points over the small track length can be considered to have the ^
An 
identical deterioration rate which continuously changes. But for  
 ^ l ^l 
short periods, the deterioration rate can be considered to be
 a n ,0  a n , p 
practically constant. Practicing engineers suggest that T
deterioration rate will be nearly constant for three to four An , x   a1n, x , , anl , x  , wherein T denote the matrix
months for tracks under high wheel loads and for four to six transposition operator.
T
months for tracks under low wheel loads. ^  ^1 ^l 
An , x   a n, x , , a n, x 
Anl   anl ,0  anl , p 

3 Copyright @ 2013 by ASME


^ l ^l ^l  in Fig. 2. The newly determined deterioration rate k n 1 is used
An   a n ,0  a n , p 
to make predictions with Eq.(3). In this stage, the deterioration
sn the average of measurement values of ti over all rate over the period from tn  r 1 to tn 1, p is considered to be
sampling points in the concerned track length on equal.
tn , x . ^
An 1  An 1,0  k n 1  Vl  X (3)
kn the degradation rate of ti within an equal
a
1
deterioration rate period including the date of the sn  r 1 n 1,1

nth inspection. t n  r 1

 1,1,
T
Vl 
 
,1 tn 1,1
 l  sn  r  2 a n 1,1
l

X  [1, 2, , p ] t n  r 1
k n 1
Modeling
a n 1, p
1
Practical knowledge of the track condition deterioration
introduced in the previous section shows that the deterioration
rate kn of ti at tn ,0 can be worked out according to
tn 1, p
{sn  r , sn  r 1 ,..., sn } and {tn  r , tn  r 1 ,..., tn } wherein r is a n 1, p
l
s n 1
determined by wheel loads conditions and inspection t n 1
frequencies of track geometry car, and for approximating the
FIG. 2. TRACK CONDITION PREDICTION DEMONSTRATION
real deterioration rate as accurately as possible the calculated
AFTER THE ( n  1) th INSPECTION
deterioration rate kn should better be adjusted once new
measurement data is available. Accordingly, the track condition As the track geometry car continues to inspect the track of
prediction model is built as follows. interest, a family of equations like Eq. (2) or (3) can be
developed, as shown in Eq.(4). Eq.(4) is the prediction model
When the nth inspection has been completed just,
for amplitudes of the irregularity parameter ti over sampling
{sn  r , sn  r 1 ,..., sn } and {tn  r , tn  r 1 ,..., tn } are used to estimate
points within the track section of concern. According to the two
the deterioration rate kn . And then kn is used to predict future
track condition assessment methods, using predictions of the
^l
track conditions a n , x with Eq.(2). The entire graphical amplitudes of all track irregularity parameters, the future local
prediction process is illustrated in Fig. 1 on which the condition and track section condition can be available in
deteriorate rate over the period from tn  r to tn , p is considered advance.
^

to be equal. A2  A2,0  k2  Vl  X 

^
An  An ,0  k n  Vl  X (2) 
^  (4)
a An  An ,0  kn  Vl  X 
1


n ,1
sn  r
tn  r  
tn ,1 ANALYSIS OF TRACK CONDITION PREDICTIONS
a n ,1
l
sn  r 1
tn  r 1
The university at which the first three authors are working
kn
has been collaborating with several bureaus of China Railroads
for a long time, such as Jinan, Kunming, and Urumqi.
a n , p
1

Therefore, we are entitled to these bureaus’ various inspections


data, e.g., inspection data of track geometry car.
tn , p It is worth noting that all collected inspection data has been
a n , p
l
sn processed with our mileage error reduction models [9, 10]. A
tn
small portion of inspection data of track geometry car over the
FIG. 1. TRACK CONDITION PREDICTION DEMONSTRATION Jiulong-Beijing railroad track section from K612+000 to
AFTER THE nth INSPECTION K614+000 is used to validate the track condition prediction
model. The validation can be conducted either by local
When a new inspection, i.e. ( n  1)th , has been completed, condition predictions or by track section predictions. Here, the
the deterioration rate kn is adjusted to be k n 1 according to error analyses on track section predictions are presented. In this
{sn  r 1 , sn  r  2 ,..., sn 1} and {tn  r 1 , tn  r  2 ,..., tn 1} as demonstrated section, two irregularity parameters’ predictions after 6

4 Copyright @ 2013 by ASME


inspections are presented, gauge and right surface. These 6 correlation coefficient is 0.9788. From these statistics, it can be
inspections span about an half year and are September 8, concluded that predicted gauge condition is accurate to plus or
September 24, October 10, October 30, November 13, and minus 0.1500mm ( ≈2*0.0739) in a probability of 0.95.
December 12, 2008, respectively. Errors in predictions on Surface
September 24, October 10, October 30, November 13,
December 12, and December 25, 2008 are analyzed. Also the three statistics are calculated for right surface
predictions and are listed in Table 3.
Gauge
From Table 3, it is seen that the maximum absolute mean
Three statistics are derived from gauge predictions, error falls on December 12 and is 0.0111mm, the maximum
including mean error (ME), standard deviation (SD), and the standard deviation is observed for predictions on September 24
correlation coefficient between predicted and actual gauge and is 0.0290mm, and the minimum correlation coefficient
standard deviations. They are tabulated in Table 2. occurs for September 24th’s predictions and is 0.9981. From
Table 2 shows that the maximum absolute mean error is these statistics, it can be concluded that predicted right surface
observed for predictions on October 30 and is 0.0330mm, the condition is accurate to plus or minus 0.0600mm (≈2*0.0290)
maximum standard deviation is 0.0739mm, and the minimum 95 percent of the time.

TABLE 2. ERROR STATISTICS FOR GAUGE PREDICTIONS


Statistics
Date ME SD 
(mm) (mm)
Sep. 24 0.0231 0.0486 0.9927
Oct. 10 -0.0006 0.0061 0.9999
Oct. 30 0.0330 0.0739 0.9788
Nov. 13 -0.0096 0.0091 0.9997
Dec. 12 0.0131 0.0190 0.9985
Dec. 25 -0.0052 0.0086 0.9997

TABLE 3. ERROR STATISTICS FOR RIGHT SURFACE PREDICTIONS


Statistics
Date ME SD 
(mm) (mm)
Sep. 24 -0.0041 0.0290 0.9981
Oct. 10 -0.0082 0.0074 0.9999
Oct. 30 0.0073 0.0130 0.9996
Nov. 13 0.0047 0.0103 0.9998
Dec. 12 -0.0111 0.0157 0.9995
Dec. 25 -0.0034 0.0060 0.9999

From Table 3, it is seen that the maximum absolute mean From predictions of the presented model, both local track
error falls on December 12 and is 0.0111mm, the maximum condition and track section condition are available two or three
standard deviation is observed for predictions on September 24 months in advance. Analyses for errors in track condition
and is 0.0290mm, and the minimum correlation coefficient predictions were made. Statistics from analyses results
occurs for September 24th’s predictions and is 0.9981. From demonstrate that predicted track condition approximates actual
these statistics, it can be concluded that predicted right surface track condition fair accurately.
condition is accurate to plus or minus 0.0600mm (≈2*0.0290)
ACKNOWLEDGEMENTS
95 percent of the time.
This research was partly sponsored by the National Key
CONCLUSIONS
Technology R&D Program under the grant of 2009BAG12A10
Condition based maintenance has proved to be an effective and the National Basic Research Program of China under the
infrastructure maintenance strategy in many industries. grant of 2012CB725406.
However, it has not been applied to railroad track maintenance
REFERENCES
to our knowledge. Condition based track maintenance relies
heavily on the availability of future track condition. This article [1] International Heavy Haul Association (IHHA), 2009,
presented a novel track condition prediction model which is “Guidelines to Best Practices For Heavy Haul Railway
based on practical knowledge of track condition deterioration. Operations- Infrastructure Construction and

5 Copyright @ 2013 by ASME


Maintenance Issues”, International Heavy Haul
Association, Virginia Beach, Virginia, USA.
[2] The Ministry of Railways of People’s Republic of China
(MOR), 2010, “Railway Line Maintenance Regulations”,
China Railway Press, Beijing, China.
[3] Esveld, C., 2001, “Modern Railway Track, 2nd edition”,
Delft University of Technology Publishing Services, Delft,
The Netherlands.
[4] CEN, 2004, “prEN13848-1:2003 Railway Applications-
Track-Track Geometry Quality-Part 1: Characterization
of Track Geometry”, CEN, Brussels, Belgium.
[5] Yoshihiko, S., 2001, “New Railway Track Dynamics”,
Beijing, China Railway Press, Beijing, pp. 98-194.
[6] Xu, P., 2012, “Mileage error correction model for track
geometry data from track geometry car & track
irregularity prediction model”, Ph.D. these, Beijing
Jiaotong University, Beijing, China.
[7] Infrastructure Inspection Center of MOR, Unpublished
results, “Track Inspection Technologies & Operation and
Maintenance of Track Inspection Car”, Infrastructure
Inspection Center of MOR, Beijing, China.
[8] Xu. P., Liu, R., Wang, F., Sun, Q., Teng, H.(Harry), 2011,
“A Novel Description Method for Track Irregularity
Evolution”, INT. J. COMPUT. INT. SYS., 4(6), pp.1358-
1366
[9] Xu, P., Sun, Q., Liu, R., Wang, F., under revision “Key
Equipment Identification Model for Correcting Milepost
Errors of Track Geometry Data from Track Inspection
Car”, TRANSPORT. RES. C-EMER.
[10] Xu, P., Sun, Q., Liu, R., Souleyrette, R., Rose, J., under
revision, “An Optimization Model for Aligning Track
Inspections of Track Geometry Car”.

6 Copyright @ 2013 by ASME

View publication stats

Anda mungkin juga menyukai