Abstract – The importance of railway transportation has
been increasing in the world. Considering the current and
future estimates of high cargo and passenger transportation
volume in railways, prevention or reduction of delays due to
any failure is becoming ever more crucial. Railway turnout
systems are one of the most critical pieces of equipment in
railway infrastructure. When incipient failures occur, they
mostly progress slowly from the fault free to the failure state.
Although studies focusing on the identification of possible
failures in railway turnout systems exist in the literature,
neither the detection nor forecasting of failure progression
has been reported. This paper presents a simple state-based
prognostic method that aims to detect and forecast failure
progression in electro-mechanical systems. The method is
compared with Hidden Markov Model based methods on
real data collected from a railway turnout system. Obtaining
statistically sufficient failure progression samples is difficult
considering that the natural progression of failures in
electro-mechanical systems may take years. In addition,
validating the classification model is difficult when the
degradation is not observable. Data collection and model
validation strategies for failure progression are also
presented.
Index Terms—Fault Diagnosis, Diagnostic expert system,
Failure Analysis, Rail transportation maintenance,
Forecasting, Prognostics, Remaining useful life estimation,
Railway Turnouts, Time Series
N
OTATION
: Constant in exponential degradation model
: Random variable following s-normal distribution
β
µ
: Mean of
2
β
σ
: Variance of
: HMM Model
: Initial probabilities of HMM
l
ji
A
,
: Transition probability from state
to
in level
l
:
Observation probability distributions
CH
: Calinski-Harabasz cluster validity index
c
n
: Number of samples clustered in cluster c
c
z
: Center of cluster c
: Center of all clusters
i
x
: Time series data sample i
k
: Number of clusters
I. I
NTRODUCTION
It is now an obligation to improve reliability,
availability, and safety of railway systems to
accommodate increasing passenger and cargo
transportation with higher train speeds, greater axle loads,
and increased service frequency. According to a report of
the European Commission, the passenger and cargo
transportation volume in European railways are expected
to double and triple respectively in 2020 [1]-[5]. It is
obvious that this demand increase cannot be satisfied with
only building new railways; the efficiency of existing and
new railways should be increased. This could be achieved
with minimum cost by increasing the availability of
railways, which is directly related with repair and
maintenance frequency.
Britain’s railway infrastructure operator, Network Rail,
was responsible for approximately 14 million minutes of
train delay in 2002-2003, costing approximately 560
million GBP [5]. Railway turnout systems are the main
component of railway infrastructure that affects the
availability of the system [6]. For example in England, 3.4
million GBP is spent every year for the maintenance of
turnout systems for 1000 km of railways [6]. Condition-
Based Maintenance (CBM), in which the health of the
machine is observed in real time and maintenance
decisions are based on the current and forecasted machine
health, increases system availability, reliability and safety
while reducing operating and support costs [7]-[8]. Thus,
the application of CBM on railway turnout systems has
critical importance for increasing the efficiency of
railways.
Diagnostics and prognostics are the two main
components of CBM [7]. Diagnostics, the process of
identifying the failure, is basically a classification
problem; prognostics, the prediction of the failure time of
the component with an incipient failure, is a forecasting
problem. Various diagnostic methods have been presented
in the literature for many diverse industrial systems [9] -
[14]. Diagnostic methods on various components of the
railway and train have also been reported [15]-[20].
There are three main approaches in the literature for
diagnostics on turnout systems: feature-, model- and
empirically-based methods. In the first approach, special
features that help to identify the failures are defined. For
example in [21], three features (i.e., irregularity, location
of maximum point, and symmetry) are defined in the time
series data. These features are obtained by analyzing the
absolute difference between the reference and actual
signals. The reference signal is collected from a fault-free
system. The absolute difference is in the form of a time
series-like reference. Irregularity is defined as an
unexpected bump in the absolute difference. The location
A Simple State-Based Prognostic Model for
Omer F. Eker, Fatih Camci, Member, IEEE, Adem Guclu, Halis Yilboga, Mehmet Sevkli, Saim
Manuscript received December 25, 2009; revised received March 26,
2010. First published XXXX; current version published XXXX. This
research was supported by The Scientific and Technological Research
Council of Turkey (TUBITAK) under project number 108M275.
Omer Faruk Eker, Adem Guclu, and Halis Yılboga are students at Fatih
University, Istanbul Turkey. (omerfarukeker@hotmail.com,
ademguclu@live.com, halis65@gmail.com). Fatih Camci is with Meliksah
University, Kayseri Turkey (fcamci@meliksah.edu.tr). Mehmet Sevkli and
Saim Baskan are with Fatih University, Istanbul, Turkey.
(msevkli@fatih.edu.tr, sbaskan@fatih.edu.tr). Copyright (c) 2009 IEEE.
Personal use of this material is permitted. However, permission to use this
material for any other purposes must be obtained from the IEEE by sending
a request to pubs-permissions@ieee.org.
IEEE Transactions on Industrial Electronics, 2010, Vol 58, Issue 5, 1718-1726