INTERNATIONAL JOURNAL OF RAIL TRANSPORTATION

https://doi.org/10.1080/23248378.2019.1703838

A time-based track quality index: Melbourne tram case study

Amir Falamarzi, Sara Moridpour and Majidreza Nazem
Civil and Infrastructure Engineering Discipline, School of Engineering, RMIT University, Melbourne, Australia

ABSTRACT ARTICLE HISTORY

Track quality indices can be used as an indicator of rail condition Received 5 May 2019
concerning the risk of damage or failure. Previous studies have Revised 9 December 2019
mainly focused on conventional rail track quality indices and light Accepted 10 December 2019
rail tracks have not been addressed properly. In order to fill this gap, KEYWORDS
this research aims to develop a track quality index which is usable Track quality index;
for both conventional and tram rail tracks. In this research dataset of degradation; geometry
the Melbourne tram network is used. In this research, based on the deviation; maintenance;
statistical analysis, track geometry parameters which are statistically Melbourne tram network
significant in the development of the proposed index for the
Melbourne tram network are determined. For the evaluation, the
predictability performance of the index proposed in this paper is
compared with the three major indices in the literature. According
to the results of the case study evaluation, the current values of the
proposed index has reasonable correlations with its previous
values.

1. Introduction
In recent decades, because of problems associated with private transportation such as
traffic congestion and environmental pollution, public transport has been encouraged
and promoted by transport authorities and urban decision-makers. Of the public trans-
port systems, rail systems are more efficient as they have higher speed and less conflict
with traffic flow. With the increase in rail transport demand, the rail transport infra-
structure is under more pressure and load. In other words, more usage of public
transport services results in higher degradation rates of tracks and rail infrastructure.
Therefore, to keep rail transport systems safe, comfortable, and reliable, the implementa-
tion of adequate maintenance and rehabilitation policies is essential [1,2]. These policies
are required to prevent rail transport systems from experiencing failures that can cause
massive human casualties and substantial financial losses.
In this context, to increase the sustainability of rail transport, railway infrastructure
maintenance management systems have been developed to optimize maintenance and
renewal activities. These infrastructure maintenance management systems cover differ-
ent tasks. The major tasks include rail monitoring and inspection, rail track degradation
prediction modelling, and short-/long-term maintenance strategy development. It is
notable that track inspection is the main prerequisite for the other tasks. Track inspection

CONTACT Amir Falamarzi [email protected] Civil and Infrastructure Engineering Discipline, School of
Engineering, RMIT University, Melbourne, Australia
© 2019 Informa UK Limited, trading as Taylor & Francis Group
2 A. FALAMARZI ET AL.

provides necessary data and information to develop rail track degradation prediction
models which can then be applied to introduce different maintenance strategies [3].
Track condition can be determined based on two different inspection types: track
geometry inspection and track structural inspection. Track structural inspections,
which cover structural parameters (rail defects and sleeper and ballast condition), are
usually undertaken by manual inspection. Track geometry inspections, which cover the
geometric parameters of the track (gauge, alignment, profile, twist, and cross-level), are
usually conducted by special track recording vehicles [4,5]. Deviation of track geometry
parameters from the allowed limits can potentially lead to a reduction in ride comfort
quality or catastrophic failures such as vehicle derailment [6,7].
To quantify and facilitate the assessment of tram track condition, a quality index is
required. This index can be also utilized later in rail degradation prediction modelling to
represent the future condition of the rail track [8–10]. Most studies carried out for the
creation of track quality indices have targeted conventional rail track degradation, while
light rail track quality index (TQI) and its related degradation models are neglected. In
this research, successful track quality indices applied and implemented in different places
are investigated. The main objectives of this study are to develop a time-based TQI based
on tram track degradation data and evaluate the performance of the proposed index. The
index proposed in this research must be comparable with the current indices. Data
derived and aggregated from the Melbourne tram network are used to elaborate the
index, and an evaluation is carried out based on the comparison of various indices with
the same dataset.
In this paper, we first present the existing literature on rail track quality indices.
Different indices and their formulation are examined. Section 3 describes the develop-
ment of a time-based TQI based on tram track geometry parameters. Section 4 describes
a case study which includes Melbourne tram track dataset. Section 5 describes the
evaluation and assessment of the proposed index. Finally, Section 6 provides the con-
clusion of this study and recommended directions for future research.

2. Literature review
Various studies have been conducted in the field of railway track degradation modelling.
However, few studies have attempted to develop and design track quality indices. Track
quality indices as track quality representatives are important as they are mostly created
based on the overall condition of tracks in terms of different track geometry parameters
[11]. In this section, different track quality indices based on track geometry parameters
used in different researches are discussed. Track geometry parameters can be categorized
into gauge, twist, alignment, profile, and cross-level. Track gauge is the right-angle
distance between two rails at a given location, below the top surface of the railhead.
A gauge defect is a deviation from the prescribed value. Track twist is the numerical
difference between two cross-levels measured at a predefined distance of a rail track
apart. Profile is the change in elevation in a specific chord length. Alignment is the
difference between the actual horizontal alignment and the designated alignment. Cross-
level is the difference between the top surfaces of two rails at a specific location [12].
Different approaches and mathematical equations have been applied to formulate
track quality indices. For instance, a synthetic track quality coefficient is used in Poland
 INTERNATIONAL JOURNAL OF RAIL TRANSPORTATION 3

to evaluate the track geometry condition based on the standard deviations (SDs) of
different track geometry parameters. The proposed index is expressed by [13,14]:
Sz þ Sy þ Sw þ 0:5Se
J¼ (1)
3:5
where J is the proposed track quality index, Sz represents the SD of profile, Sy denotes the
SD of alignment, Sw is the SD of track twist, and Se represents the SD of track gauge.
The chord length (representing the length of measurement for collecting track geo-
metry parameters) of 10 m is used in this study. The permissible values for the Polish
index based on train speed are presented in Table 1.
In the USA, a track roughness index was established by Amtrak to represent the
condition of rail tracks. This index can be calculated by the average of squared differential
geometry deviations over a chord length of 20 m as follows [15,16]:

1Xn1
r2 ¼ ðGdevtþ1  Gdevt Þ2 (2)
n i¼1

where r2 represents track roughness value, n is the number of measurements, and Gdevtþ1
and Gdevt represent the amount of gauge deviation for two consecutive years.
The proposed index can be utilized for gauge, cross-level, alignment, and profile. Table 2
demonstrates the condition of tracks associated with the roughness index.
The Canadian National Railway Company (CN) has introduced a TQI based on the
squared SDs of track geometry parameters as follows [17,18]:

TQIi ¼ 1000  C  σ 2i (3)

where TQI represents the proposed track quality index, C is constant and is determined
as 700 for main track lines, and σ i is the SD of track geometry parameters.
The above equation can be used for assessing individual track geometry parameters. In
order to calculate overall TQI (OTQI), the average of indices for track geometry para-
meters has been proposed by CN. Accordingly, a smaller TQI indicates that the track
segments are at a greater risk of failure.
In China, the Chinese national railroads use the sum of SDs of total track geometry
parameters for calculating the OTQI as follows [19,20]:
X
n
TQI ¼ σi (4)
i¼1

Table 1. Allowable values of J index for different speeds.

Speed (km/h) 30 40 90 120 160 200
J index 12.0 11.0 6.2 4.0 2.0 1.4

Table 2. Condition of track in accordance with the roughness index.

r2 value r2 < 1 1 < r2 < 4 4 < r2 < 16 r2 < 16
Track condition Very good Good Average Poor
4 A. FALAMARZI ET AL.

where TQI represents the track quality index and σ i is the SD of track geometry
parameters.
In this research, two different lengths for assessing overall track quality are proposed.
For high-speed railroads, a track length of 500 m is applied, and for conventional
railroads, a track length of 200 m is applied. Larger TQI values imply a potential
reduction in overall track quality.
Railway track geometry parameters can be assessed according to European Standard
EN 13848-5. This standard can be applied to profile, alignment, and gauge parameters.
To calculate this index, the SDs of the alignment and profile are used. For the gauge
parameter, the difference between the mean value of the gauge and the pre-defined gauge
value is applied. In Table 3, recommendations of EN 13848-5 for profiles based on track
quality classes (TQCs) and two specific speed categories are shown. The TQC ‘A’
represents the safest condition of a rail track, and the TQC ‘E’ represents the critical
condition of a rail track. In Table 4, the thresholds for assessment of gauge parameter
based on the difference values and two specific speed categories over a length of a 100-
m track segment are tabulated [21].
In India, the national railway uses an exponential equation to calculate the TQI. The
index is defined for individual track geometry parameters (e.g. gauge and profile) as
follows [22,23]:

GI ¼ 100  eðSDm SDN Þ=ðSDU SDN Þ (5)

where GI is the individual track geometry index (TGI), SDm represents the SD of the
measured parameter, SDU denotes the SD value of a track with immediate maintenance
need, and SDN indicates the SD of a newly laid track. The values of SDU and SDN are
constant and can be found in Table 5.
The overall TGI (OTGI) can be obtained from:
2UI þ TI þ 6AI þ GI
TGI ¼ (6)
10
where TGI represents the overall track geometry index, UI is the index for unevenness,
TI indicates the index for twist, AI denotes the index for alignment and GI is the index
for gauge, which can be calculated using Equation 5. The allowable amounts of TGI are
shown in Table 6.

Table 3. Recommendations of EN 13848 for assessment of profile.

Threshold of SD (mm)
TQC
Speed (km/h) A B C D E
80 < V ≤ 120 <0.75 1.10 1.80 2.50 >2.50
120 < V ≤ 160 <0.65 0.85 1.40 1.85 >1.85

Table 4. Recommendations of EN 13848 for assessment of gauge.

Difference between the mean value of gauge and predefined gauge value
Speed (km/h) Alert limit (SL) Intervention limit (IL) Safety limit (AL)
80 < V ≤ 120 −7≤ difference ≤27 −6≤ difference ≤25 −5≤ difference ≤22
120 < V ≤ 160 −5≤ difference ≤20 −4≤ difference ≤18 −3≤ difference ≤16
 INTERNATIONAL JOURNAL OF RAIL TRANSPORTATION 5

Table 5. The values of SDN and SDU based on the track chord length and speed.
SDU
Parameters Segment length (m) SDN (mm) Speed > 105 km/h Speed < 105 km/h
Alignment 7.2 1.5 3 3
Gauge - 1 3.6 3.6
Twist 3.6 1.75 3.8 3.8
Unevenness 9.6 2.5 6.2 7.2

Table 6. Condition of track based on TGI values.

TGI value TGI<36 36< TGI<50 50< TGI<80 80< TGI
Track condition Poor Average Good Excellent

The Swedish national rail network uses its own TQI to assess track geometry
condition. This index is based on the SDs of track parameters and is obtained
from [24,25]:
h i
100 σσHH þ 2  σσS S
Q ¼ 150  lim lim
(7)
3
whereQ represents the index for evaluating track geometry condition, σ H is the
average of SDs of left and right profiles, σ S denotes the SDs of other track
geometry parameters including gauge, cross-level, and horizontal deviation,
σ Hlim indicates the allowable limit of σ H , and σ Slim indicates the allowable limit of
σ S based on track type.
According to Swedish standards, the maximum and minimum values for the
proposed index are 50 and 150, respectively. The allowable values of the index are
in the range of 70–90. The chord length of 12 m is used to measure the
deviations.
In Iran, based on SD values and the mean values of the track geometry parameters,
TGIs for individual parameters and OTGI have been proposed. A chord length of 19 m is
used in that research for collecting the deviation values of parameters. These parameters
include track gauge, alignment, profile, and twist. The following equation is used for
assessing the alignment parameter [11]:
jxAlignLeft j þ 3  SDAlignLeft þ jxAlignRight j þ 3  SDAlignRight
AI ¼ (8)
2
where AI represents the value of the index. xAllignLeft and xAllignRight represent the mean
value of alignment for left and right rails. SDAllignLeft and SDAllignRight represent the SD of
alignment for left and right rails.
For overall track geometry assessment, the following formula has been
proposed:
a
 GI þ þ a2  GI  þ b  AI þ c  PI þ d  TI
OTGI ¼ 2 (9)
2 þbþcþd
aþa

where OTGI represents the overall TGI, GI þ is the positive gauge index, GI  indicates
the negative gauge index, AI, PI, and TI represent, respectively, the alignment, the profile,
6 A. FALAMARZI ET AL.

Table 7. Allowable values of OTGI based

on track class.
Track class Allowable values of OTGI
A 2.19< OTGI <4.91
B 3.02< OTGI <6.26
C 3.62< OTGI <7.23
D 7.06< OTGI <8.70

Table 8. Condition of track based on w5 .

Track condition w5 values
New tracks w5 < 0.1
Good condition 0.1 <w5 < 0.2
Average condition 0.1 <w5 < 0.6
Poor condition w5 > 0.6

and the twist indices, a, a, b, c, and d are constant parameters which vary between 0.08
and 1.00, based on track class and the number of defects in a certain chord length.
Table 7 shows the allowable values of OTGI for different track classes and the
maximum of two defects.
The Austrian Railway has proposed a track defectiveness index based on the ratio
between the total length of segments that have exceeded the acceptable limit and the total
length of the track. According to this method, a defectiveness index for each track
geometry parameter can be calculated using the following formula [13]:
P
Li
w¼ (10)
L
where w is the defectiveness index of geometry parameters, Li represents the sum of the
length of segments that exceed the acceptable range and L denotes the total length of
track segments.
Larger values of the track defectiveness index represent a reduction in track quality.
The overall track geometry defectiveness index (five-parameter index) can be calcu-
lated by:
  
w5 ¼ 1  ð1  we Þ 1  wg 1  wy ð1  wz Þð1  ww Þ (11)

where w5 is the five-parameter defectiveness index, wg represents the defectiveness of

cross-level, we is the defectiveness of gauge and wy denote the arithmetic average of
profile, wz and ww , respectively, denote the arithmetic average of alignment and twist.
The Austrian Railway has provided Table for track condition assessment based on the
overall track geometry defectiveness index (Table 8):
In this section, different track quality indices have been investigated and reviewed.
Important parameters applied in development and formulation of track quality indices
can be summarized into SD of deviations of track geometry parameters, the mean value
of deviations of the parameters, average differential deviations of the parameters, and the
length of segments which are prone to deviation and defectiveness. These indices along
with their thresholds can be utilized by track maintenance systems to assess the condition
of track geometry parameters.
 INTERNATIONAL JOURNAL OF RAIL TRANSPORTATION 7

3. Development of tram track quality index

Quality indices for rail track represent the quality of rail tracks with regard to various
track geometry parameters. The proposed index should be easy to use and, at the same
time, reflect the geometry condition of track segments properly. Based on the findings
obtained from the literature review, two main parameters including the mean value of
geometry deviation in different measurements (years) and also the average differential
geometry deviation have a strong contribution in the development of track quality
indices. These parameters have been used and implemented separately in different
quality indices.
The mean value of the geometry deviations is an essential factor for formulating the
quality index, as larger values demonstrate more deviation from the original geometry
parameter and consequently more risk of track failure compared to rail tracks with the
lower mean values. Furthermore, the role of the average differential geometry deviation is
important. It is notable that two different track segments can have an accidentally equal
mean value of geometry deviation, but the one with a larger differential geometry
deviation reflects a faster rate of degradation in track geometry than the segment with
a lower differential geometry deviation. On the other hand, two different track segments
may have an equal average differential geometry deviation, but the one with a higher
mean value represents a faster rate of degradation.
In this research, with regard to the importance of the above parameters, a new TQI has
been proposed based on the combination of them. By utilizing these parameters, the
proposed index can benefit from the advantages of each of them. The proposed TQI is
formulated as follows:

TQIi ¼ μi þ λi (12)

whereTQIi represents the tram track quality index based on the geometry deviation
values including Gauge Deviation (GD), Twist Deviation (TD), Alignment Deviation
(AD), Profile Deviation (PD), and Cross-level Deviation (CD) for the track segment i. μi
is the mean value of the geometry deviations for the track segment i and λi represents the
average differential geometry deviation for the track segment i.
The mean values of the geometry deviation of track segments for the consecutive years
(μi Þ can be calculated using the following formula:

1X m
μi ¼ Gdevt (13)
m t¼1

whereμi is the mean value of the geometry deviations for the track segment i, Gdevt
represents the geometry deviation of track segment i in year t, and m denotes the number
of years for which data were collected.
The average differential geometry deviation (λi Þ was also included in the index
formulation. λi can be determined by dividing the sum of the absolute value of differences
between two consecutive geometry deviations by the total number of data collection years
for the track segment i as follows:
8 A. FALAMARZI ET AL.

1Xm1
λi ¼ jGdevtþ1  Gdevt j (14)
m t¼1

where λi represent the average differential geometry deviation of the segment i, and Gdevt
and Gdevtþ1 represent two consecutive geometry deviation values for the track segment i.
Once the values of TQIi are calculated individually for all the track geometry para-
meters, then the Overall Track Quality Index (OTQI) can be obtained. The value of
OTQI can be calculated by using the following formula:
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
a  TQIGD 2 þ b  TQITD 2 þ c  TQIAD 2 þ d  TQIPD 2 þ e  TQICD 2
OTQIi ¼
aþbþcþdþe
(15)
where OTQIi represents the overall track quality index for the segment i, TQIGD , TQITD ,
TQIAD , TQIPD , and TQICD denote TQI values based on the GD, TD, AD, PD, and CD,
respectively, for the track segment i, and a, b, c, d, and e are constant coefficients. The
constant coefficients equal to 1, while the geometry parameters associated with each of
these coefficients play a role in the development of the index. Otherwise, these coeffi-
cients will be zero.
As discussed above, regarding the number of track geometry parameters involved in
a certain dataset, Equation 15 can be modified.
In the next sections, based on the dataset of the Melbourne tram network, the
proposed index is implemented and assessed.

4. Case study
In this research, the Melbourne tram network, which comprises over 1700 stops, 25
routes, and 250 km of double tracks, is considered as the case study. The Melbourne tram
system is one of the largest and busiest tram networks in the world. In 2017, by operating
450 in-service tram cars, more than 204 million passengers were carried by the tram
system of Melbourne, which shows 400,000 increase in patronage compared with 2016.
The Melbourne tram network has the highest light rail ridership in Australia [26]. The
dataset for this study was provided by Yarra Trams, the operator of the Melbourne tram
network [27]. The dataset consists of different track types, including curves, straights,
crossovers, and H-crossings, and covers track geometry parameters including gauge,
twist, cross-level, and profile and horizontal alignment, along with traffic volumes and
other rail parameters such as track surface, rail profile, rail type, and rail support. The
collected data cover six sequential years (2010–2015). Note that in this research a chord
length (i.e. the length of the measure for collecting track geometry parameters) of
10 m was applied.
In this study, in order to extract the geometry deviations of track geometry parameters
associated with several years, data segmentation technique should be applied. Data
segmentation is the process of converting track record data into track segments to
facilitate the process of data matching. In this research, align with the chord length,
a length of 10 m was selected for development of a track segment. Each track segment
represents a track record with a specific identification code (a combination of track code
 INTERNATIONAL JOURNAL OF RAIL TRANSPORTATION 9

and details of measurement location). By using this identification code, the deviation
values of the geometry parameters for six consecutive years of each track segment were
collated and a multi-year dataset was built. In this regard, more than 34,000 track
segments have been processed and analysed.
Basically, for the establishment of a degradation model as well as quality index, the
existence of a meaningful correlation between previous and existing values of track
geometry parameters is required [28]. Involving parameters which are not statistically
significant can reduce the accuracy of the proposed index. In this study, the Pearson
correlation test as a measure of the relationship strength between two numeric variables
has been used for both existing and previous values of the geometry parameters. The
Pearson correlation coefficients were calculated between −1 and 1. Coefficients closer to 1
and −1 have a greater correlation with the target variable and 0 means no correlation.
Track geometry parameters which are correlated acceptably can be involved in the
development of degradation models. Otherwise, they are not statistically significant
and should be removed from the index development process. As demonstrated in
Table 9, significant correlations between previous and existing deviations of gauge and
twist parameters have been identified. The rest of the parameters is not useful in the
development of the proposed index.
Before using the dataset for further processes, data filtration technique has been done
to remove error and out-of-range data in order to increase the accuracy of the proposed
index. A sample of changes in track gauge deviation of track segments over travelled
distance for a particular track section is illustrated in Figure 1.
For data filtration, determining the distribution patterns of the current dataset is
useful. If the distribution of the dataset matches a normal distribution, 99.7% of the
data will be within the distance of 3  SD from the mean value [29]. For this purpose, the
Shapiro–Wilk tests were conducted. This test is useful measures to check the possibility
of a normal distribution.
After analysing different track sections, it was determined that the changes in GD and
TD values mainly followed a normal distribution (Shapiro–Wilk test: p-value >0.05). In
Figure 2, the frequency histogram for a certain track section which follows a normal
distribution is illustrated. In this section, track segments deviating from μ  3  SD were
identified and replaced with the closest values.
In this study, the conditions of rail track were examined concerning GD (for both positive
gauges, where railheads diverge from the centreline of the track, and negative gauge, where
railheads converge towards the centreline of the track) and TD. Consistent with the literature
[30–32] and based on the GD and TD values, the rail conditions were classified into different
levels, and the values associated with TQI were calculated. Table 10 presents the track
condition based on the TQI values for the gauge parameter, and Table 11 presents the track

Table 9. The results of the Pearson correlation analysis.

Geometric deviations Pearson corr. coefficient p-Value
Gauge 0.87 0.00
Twist 0.80 0.00
Alignment 0.17 0.00
Profile 0.13 0.00
Cross-level 0.25 0.00
10 A. FALAMARZI ET AL.

4
3
2
GD (mm)
1
0
-1
-2
-3
20 120 220 320 420 520 620 720 820 920
Distance (meter)

Figure 1. Changes in gauge deviation of track segments for a particular track section.

10
9
8
7
Frequency

6
5
4
3
2
1
0
-2.5 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5 3
GD (mm)

Figure 2. Frequency histogram for a certain track section.

condition based on the TQI values for the twist parameter. Finally, Table 12 presents the
overall track condition based on the OTQI values.
In Tables 10–12, IAL or Immediate Action Limit denotes the value if exceeded, train
speed restrictions or prompt correction of track geometry will be required. IL or
Intervention Limit denotes the value if exceeded, corrective maintenance operations
are required to avoid IAL. AL or Alert Limit denotes the value if occurs, track geometry
condition should be analysed and the planned maintenance operations are required
regularly.
According to Table 12, for example, if OTQI values for positive GD are ranged
between 14.5 and 17.5, planned maintenance operations should be scheduled. If OTQI
values for positive GD are ranged between 17.5 and 21.5, corrective maintenance opera-
tions should be scheduled. Lastly, if OTQI values exceed 21.5, prompt correction of rail
track is required.
In order to better understand how the proposed indices can be used to determine the
condition of rail tracks regarding failure risk, a sample track section with the length of
50 m (containing five track segments) from the case study was examined. In this context,
μi and λi were calculated using Equation 13 and Equation 14, respectively. TQI (Equation
 INTERNATIONAL JOURNAL OF RAIL TRANSPORTATION 11

Table 10. Track condition based on TQIGD.

GD limit TQI
Positive GD IAL 35≤ TQI
IL 30≤ TQI <35
AL 25≤ TQI <30
Negative GD IAL TQI ≤−11
IL −11≤ TQI <−9
AL −9≤ TQI <−7

Table 11. Track condition based

on TQITD.
Limit TQI
IAL 24.5≤ TQI
IL 17.5≤ TQI <24.5
AL 14≤ TQI <17.5

Table 12. Overall track condition based on OTQI.

GD Limit OTQI
Positive GD IAL 21.5≤ OTQI
IL 17.5≤ OTQI <21.5
AL 14.5≤ OTQI <17.5
Negative GD IAL OTQI ≤−13.5
IL 9.5≤ OTQI <13.5
AL 7.5≤ OTQI <9.5

Table 13. TQI and OTQI values obtained for track segments of a sample track section.
Geometry deviation (Gdev Þ values measured over 6 years
Seg. No. Geometry parameters 2010 2011 2012 2013 2014 2015 μi λi TQI OTQI
1 GD 5.61 1.76 1.69 4.97 7.38 4.95 4.39 2.01 6.40 5.32
TD 6.45 7.1 7.67 7.65 8.38 10.05 7.88 0.61 8.49
2 GD 12.07 10.72 11.92 5.69 3.2 10.21 8.97 3.05 12.02 6.34
TD 3.76 3.84 3.88 3.63 4.05 4.24 3.90 0.16 4.06
3 GD 4.81 4.76 3.85 5.65 2.92 3.97 4.33 1.09 5.42 3.32
TD 3.16 3.78 3.1 3.11 3.27 4.17 3.43 0.40 3.83
4 GD 3.49 2.77 2.07 3.59 3.07 4.23 3.20 0.77 3.97 3.47
TD 4.64 4.69 4.8 5.45 5.69 6.79 5.34 0.36 5.70
5 GD 8.75 7.32 5.51 7.59 8.88 9.95 8.00 1.28 9.28 5.17
TD 3.2 3.49 4.19 4.22 5.09 5.18 4.23 0.33 4.56

12) and OTQI (Equation 15) were then calculated and the results are tabulated in
Table 13.
According to Table 13, the TQI values calculated from the GDs range between 3.33
and 6.34. According to Table 12 and based on the calculated OTQI values, these segments
are categorized in no safety concern zone. For instance, Segment 2 has the highest OTQI,
which implies a higher risk of rail track failure in the long term compared with other
segments, and this segment should, therefore, be treated with higher priority than the
others. Segments 1 and 3 have accidentally similar mean values of gauge deviation, but as
the average differential gauge deviation of Segment 1 is higher, its TQIGD value is larger.
On the other hand, for example, Segments 4 and 5 have almost similar average differ-
ential twist deviation, but as the mean value of twist deviation of Segment 4 is higher,
12 A. FALAMARZI ET AL.

consequently its TQITD value is larger. These examples demonstrate the importance of
the average differential geometry deviation as well as the mean value of geometry
deviation in the proposed quality index.

5. Evaluation
In this section, the evaluation of the proposed index implemented on the Melbourne
tram network is presented. As discussed, the main roles of quality indices are to represent
the current condition of rail track geometry parameters of track segments as well as their
future conditions. The future value of a quality index can be used by rail organizers and
operators to address preventive maintenance strategies prior to rail track failures. One of
the potential ways to obtain the future value of a quality index is to predict it by applying
existing data. For this purpose, the predictability performance of the proposed index
should be analysed. An index with greater predictability performance can be used by
predictive models to provide forecasts more effectively. In this regard, the correlation
between the consecutive values of a TQI for a specific route or network is essential (e.g.
OTQI2015 and OTQI2014). A stronger correlation between the values of an index in
consecutive years demonstrates that the index has greater predictability performance.
In this research for carrying out the evaluation, the correlation between the current
value of OTQI index and the previous value of OTQI is examined. The current value of
the quality index is obtained based on the geometry deviations measured up to the
current year. While for the previous value of the quality index, geometry deviations are
processed up to the last year. In order to compare the performance of the proposed index
and other studied indices, this process has been carried out for them as well. Three major
quality indices including the index based on the SD, the index based on the average of
squared differential geometry deviation, and the index based on the mean value and SD
are included. To evaluate the performance of the indices, the Melbourne tram network
dataset is used. In this dataset, the values of the previous and current OTQI, as well as
other three indices for each track segments, are calculated. The Pearson correlation
analysis was applied. Figure 3 illustrates the correlation between the current and previous
values of the indices. Table 14 presents the results of the analysis in terms of the
performance indicators including the Pearson correlation coefficient and Root-Mean-
Squared Error (RMSE).
Figure 3 illustrates the correlation between the previous and current values of the indices
in the scatter plots. Indices where data points are more scattered and located with more
distances have lower correlation coefficients. According to Table 14 and based on the
Pearson correlation analysis, RMSE values associated with the indices range between 0.35
and 5.93. Also, the Pearson correlation coefficients associated with the indices range
between 0.77 and 0.97. Indices with lower RMSE and greater correlation coefficients can
provide more accurate geometry degradation predictions for the Melbourne tram network
dataset. As represented in this table, OTQI and J Index have lower RMSE compared to the
other indices. On the other hand, OTQI and OTGI indices have greater correlation
coefficients than the rest of the indices. Regarding the performance indicators, the proposed
index can be applied in tram track degradation prediction models with acceptable accuracy.
 INTERNATIONAL JOURNAL OF RAIL TRANSPORTATION 13

R2=0.95 R2=0.85

R2=0.67 R2=0.60

Figure 3. The correlation between the current and previous values of the indices.

Table 14. The results of the analysis in terms of the Pearson correlation
coefficient and RMSE.
Degradation index Pearson corr. coef. Adjusted R2 RMSE
OTQI 0.97 0.95 0.35
OTGI 0.92 0.85 1.20
J Index 0.82 0.67 0.37
Sq. diff. index 0.77 0.60 5.93

6. Conclusions and future research directions

A tram TQI is proposed in this paper based on track geometry parameters. TQI is
a useful measure for infrastructure maintenance management systems as well as
prioritizing and ranking rail track segments with maintenance needs. To formulate
the tram TQI and based on the finding of successful quality indices, the combination of
the mean value of geometry deviations and the average of absolute differential geome-
try deviations was used. The proposed index can be applied individually on a specific
geometry parameter. In this study, the OTQI was also proposed based on the different
track geometry parameters. The OTQI can examine the condition of track segments
comprehensively. In the proposed index, the segments with larger OTQI values are
more exposed to degradation and failure than their counterparts with lower OTQI
14 A. FALAMARZI ET AL.

values. In this study as a case study, a dataset collected from the Melbourne tram
network, which comprises 250 km of double rail, was analysed. Before developing the
index, outliers were replaced with the nearest values, and track sections were divided
into smaller track segments.
To evaluate and assess the accuracy of the proposed index, the predictability
performance of the proposed index along with three major indices was examined. An
index which demonstrates greater predictability performance can provide more accu-
rate forecasts when is utilized by predictive models. For this purpose, track geometry
dataset of the Melbourne tram network has been analysed. The previous and current
values of the above indices and correlation coefficients were calculated. According to
the results, the proposed index presents a reasonable correlation with acceptable
accuracy. The findings of this research indicate that the proposed index can be used
as an effective measure for the assessment of the geometric condition of tram tracks as
it is easy to develop and apply in the establishment of predictive models based on the
previous values of the index. The main benefit of the application of the proposed index
compared to other indices which have been investigated in this research is its predict-
ability accuracy.
The main novelty of this research which can distinguish it from other research is the
development and implementation of a TQI which has been developed and evaluated
based on tram track degradation data. For future research directions in association with
the TQI, consideration of other potential factors in addition to those discussed in this
study may be useful and can enrich the rail degradation prediction modelling.
Furthermore, for the evaluation of the proposed index, datasets associated with different
tram track networks along with examining different track geometry parameters in the
development of the index can be helpful.

Acknowledgements
The authors would like to note their special thanks to Yarra Trams for providing and getting access
to the datasets of this study.

Disclosure statement
No potential conflict of interest was reported by the authors.

References
[1] Sanudo R, Dell’Olio L, Casado J, et al. Track transitions in railways: a review. Constr Build
Mater. 2016;112:140–157.
[2] Gaudry M, Lapeyre B, Quinet E. Infrastructure maintenance, regeneration and service
quality economics: a rail example. Transp Res Part B Methodol. 2016;86:181–210.
[3] Falamarzi A, Moridpour S, Nazem M, et al. Rail degradation prediction models for tram
system: Melbourne case study. J Adv Transp. 2018a;2018.
[4] Weston P, Roberts C, Yeo G, et al. Perspectives on railway track geometry condition
monitoring from in-service railway vehicles. Veh Syst Dyn. 2015;53(7):1063–1091.
[5] Soleimanmeigouni I, Ahmadi A, Kumar U. Track geometry degradation and maintenance
modelling: a review. Proc Inst Mech Eng F J Rail Rapid Transit. 2016;232(1):1–30.
 INTERNATIONAL JOURNAL OF RAIL TRANSPORTATION 15

[6] Dindar S, Kaewunruen S, An M. Identification of appropriate risk analysis techniques for

railway turnout systems. J Risk Res. 2018;21(8):974–995.
[7] Jamshidi A, Núñez A, Dollevoet, et al. Robust and predictive fuzzy key performance
indicators for condition-based treatment of squats in railway infrastructures. J Infrastruct
Syst. 2017;23(3):1–13.
[8] Bai L, Liu R, Sun Q, et al. Classification-learning-based framework for predicting railway
track irregularities. Proc Inst Mech Eng F J Rail Rapid Transit. 2016;230(2):598–610.
[9] Andrade R, Teixeira P. Statistical modelling of railway track geometry degradation using
hierarchical Bayesian models. Reliab Eng Syst Saf. 2015;142:169–183.
[10] Falamarzi A, Moridpour S, Nazem M, et al. Development of a random forests regression
model to predict track degradation index: Melbourne case study. Australian Transport
Research Forum 2018; Darwin, Australia; 2018.
[11] Sadeghi J. Development of railway track geometry indices based on statistical distribution of
geometry data. J Transp Eng. 2010;136(8):693–700.
[12] He Q, Li H, Bhattacharjya D, et al. Track geometry defect rectification based on track
deterioration modelling and derailment risk assessment. J Oper Res Soc. 2015;66
(3):392–404.
[13] Madejski J, Grabczyk J. Continuous geometry measurement for diagnostics of tracks and
switches. Proceedings of the International Conference on Switches; 2002; Delft,
Netherlands.
[14] Chudzikiewicz A, Bogacz R, Kostrzewski M, et al. Condition monitoring of railway track
systems by using acceleration signals on wheelset axle-boxes. Transport. 2018;33(2):30–42.
[15] Van Der Westhuizen NJ, Grabe PJ. The integration of railway asset management informa-
tion to ensure maintenance effectiveness. J South Afr Inst Civil Eng. 2013;55(3):18–29.
[16] Liu RK, Xu P, Sun ZZ, et al. Establishment of track quality index standard recommendations
for Beijing metro. Discrete Dyn Nat Soc. 2015;2015.
[17] Setiawan D, Rosyidi SA. Track quality index as track quality indicator. Proceedings of the
19th international symposium of FSTPT; Yogyakarta, Indonesia; 2016.
[18] Roghani A. A quantitative evaluation of the impact of soft subgrades on railway track
structure [dissertation]. Edmonton (Canada): University of Alberta; 2017.
[19] Xu P, Sun Q, Liu R, et al. A short-range prediction model for track quality index. Proc Inst
Mech Eng F J Rail Rapid Transit. 2011;225(3):277–285.
[20] Li FY, Xu YD, Li HF, et al. A comparative study of the TQI method and process perfor-
mance index method in the quality evaluation of track fine adjustment. Proceedings of the
15th International Conference on Railway Engineering Design and Operation. 162; 2016;
Madrid, Spain. p. 33–40.
[21] EN 13848-5. Railway applications – track geometry quality – part 5: geometric quality
levels – plain line, switches and crossings. Brussels: European Committee for
Standardization (CEN); 2017.
[22] Mundrey JS. Railway track engineering (Fourth edition). New Delhi: Tata McGraw-Hill
Education; 2009.
[23] Berawi ARB. Improving railway track maintenance using power spectral density (PSD)
[dissertation]. Porto (Portugal): Universidade do Porto; 2013.
[24] Andrade AR, Teixeira PF. A Bayesian model to assess rail track geometry degradation
through its life-cycle. Res Transp Econ. 2012;36(1):1–8.
[25] Odolinski K, Smith AS. Assessing the cost impact of competitive tendering in rail infra-
structure maintenance services: evidence from the Swedish reforms (1999 to 2011).
J Transport Econ Policy. 2016;50(1):93–112.
[26] Currie G, Burke M. Light rail in Australia – performance and prospects. Australasian
Transport Research Forum 2013; Brisbane, Australia; 2013.
[27] Public Transport Victoria (PTV). Public transport victoria annual report. Public Transport
Victoria: Melbourne; 2017.
[28] Sadeghi J, Askarinejad H. Development of improved railway track degradation models.
Struct Infrastruct Eng. 2010;6(6):675–688.
16 A. FALAMARZI ET AL.

[29] DeGroot MH, Schervish MJ. Probability and statistics. Pearson Education; 2012.
[30] Kopf F, Maras I, Gasser F, et al. Cost-effective track maintenance, renewal & refurbishment
methods. Urban Track: Belgium; 2009.
[31] Wilson A, Kerr M. Rail installation and repair. Sydney: NSW Transport Corp; 2013.
[32] Guler H. Prediction of railway track geometry degradation using artificial neural networks:
a case study for Turkish state railways. Struct Infrastruct Eng. 2014;10(5):614–626.