Template for an Acta IMEKO event paper


ACTA IMEKO 
ISSN: 2221-870X 
December 2016, Volume 5, Number 4, 56-63 

 

ACTA IMEKO | www.imeko.org December 2016 | Volume 5 | Number 4 | 56 

CBM for a fleet of railway vehicles: infrastructure and 
algorithms 
Micaela Caserza Magro1, Paolo Pinceti1, Marco Antonelli2, Enrico De Paola2, Pierluigi Firpo2, Enrico 
Marino2 

1University of Genoa, Dept. DITEN, via all’Opera Pia 11a, 16145, Genova, Italy 
2Bombardier Transportation Italy Spa, via Tecnomasio, 17047, Savona, Italy   

 

 

Section: RESEARCH PAPER  

Keywords: railway vehicles; Condition Based Maintenance (CBM); key maintenance indicators; condition monitoring; industrial communication 

Citation: Micaela Caserza Magro, Paolo Pinceti, Marco Antonelli, Enrico De Paola, Pierluigi Firpo, Enrico Marino, CBM for a fleet of railway vehicles: 
infrastructure and algorithms, Acta IMEKO, vol. 5, no. 4, article 9, December 2016, identifier: IMEKO-ACTA-05 (2016)-04-09 

Section Editor: Lorenzo Ciani, University of Florence, Italy 

Received September 7, 2016; In final form November 9, 2016; Published December 2016 

Copyright: © 2016 IMEKO. This is an open-access article distributed under the terms of the Creative Commons Attribution 3.0 License, which permits 
unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited 

Funding: This work was supported by Regione Liguria, Project (PO CRO FSE 2007-2013 Asse IV) 

Corresponding author: Paolo Pinceti, e-mail: paolo.pinceti@unige.it 

 

1. INTRODUCTION 
Optimizing the maintenance for a fleet of machineries or 

vehicles means to guarantee a high Quality of Service (QoS) 
with a minimum number of interventions. The concept of QoS 
varies for each specific application. For railway vehicles, the 
passengers that want to have reliable trains on time measure the 
QoS. For machinery, QoS means almost often to work without 
interruptions and with a constant quality of the production. In 
all cases, the QoS is related to the availability that is the 
percentage of time a system is in normal operating conditions. 

An optimized maintenance strategy coordinates the 
scheduled  stops   with   the  maintenance    interventions,   e.g.  

 

 
 
 

railway vehicles must stop for safety reasons a fixed number of 
times per year or after a given number of kilometres according 
to national regulations and laws. Maintenance policies are 
discussed in [1]-[9]. 

Condition Based Maintenance (CBM) and Predictive 
Maintenance (PM) seem to be the most effective policies, since 
their goal is to start a maintenance intervention only when it is 
necessary. To do this, it is mandatory to monitor the operation 
of a vehicle (or a machinery, or a plant) to find symptoms of 
incoming failures (“fix it before it fails”). For this purpose, 
various metrics and sensor-based methods can be used to 
measure and monitor continuously the condition of the 

ABSTRACT  
Data acquisition and communication technologies give the possibility of receiving and storing a huge amount of data from machinery 
and plants in operation. From these data it is possible to create a set of Key Maintenance Indicators (KMI) useful for optimizing the 
maintenance policy. Raw data from the field are to be processed and filtered for obtaining effective KMIs to use in algorithms aimed at 
discovering anomalies or abnormal operation of one or more machineries or plants.  
This paper presents a roadmap towards the Condition Based Maintenance of a fleet of railway vehicles. The paper associates to each 
maintenance policy its benefits and its requirements in terms of technological infrastructure and operating costs. Bombardier 
Transportation Italy started this roadmap a few years ago, for moving from a reactive maintenance policy to a proactive policy. 
Increasing the effectiveness of maintenance implies the sensorization of the machines and the creation of a network for funneling 
information from the machineries to the central maintenance room. A Company must find an equilibrium point between complexity 
and expected benefits.  
Results achieved by means of a specifically developed tool for data analysis applied to some sub-systems of the vehicles are presented. 



 

ACTA IMEKO | www.imeko.org December 2016 | Volume 5 | Number 4 | 57 

equipment. To do this, we introduced the concept of Key 
Maintenance Indicator (KMI) that may represent: 

- a simple counter of variables of the vehicle (e.g. distance 
travelled, hours of operation, the number of times a door 
opened and closed, etc.; 

- a calculated variable that uses different parameters of the 
vehicle (e.g. number of occurrence of “X” when “Y” was on, 
etc.); 

- measures from sensors (e.g. oil analysis, vibration analysis, 
etc.) specifically installed for maintenance purposes. 

The KMIs are the foundation for the construction of 
algorithms aimed at monitoring the status of the vehicle and at 
deciding maintenance interventions [11], [12].    

Collecting and processing the raw data that will become 
KMIs require a complex technological infrastructure described 
into details in [10]. Railway vehicles require two sub-systems: 

- “on-board system” (the train): that produces and collects 
raw data and sends them to the; 

- “off-board system”: collects data from all the vehicles and 
implement the algorithms for CBM. 

This paper shortly describes the on-board and off-board 
infrastructures, and focuses on the procedures for the data 
analysis and for the discovery of rules and metrics useful for 
CBM; in other words, the “Conditions” to active a maintenance 
intervention. A joint research team of Bombardier 
Transportation Italy and of the University of Genoa, Dept. 
DITEN, developed a tool for the data filtering, sorting, and 
analysis that was applied on the historical database of about 
5000 vehicles to find rules for deciding the maintenance 
interventions. The paper presents some results of this analysis 
for three different sub-systems.   

2. THE TECHNOLOGICAL INFRASTRUCTURE  
2.1. The on-board infrastructure  

Railway vehicles are composed by a set of independent 
subsystems, each one equipped with a dedicated Train Control 
& Management System (TCMS) that collects data both for 
control purposes and for diagnosis. All the TCMSs 
communicate via the Multifunction Vehicle Bus (MVB), 
standardized in IEC 61375-3. TCMSs transmit the process data 
for the vehicle control with a sampling time of 1 second. When 
a TCMS detects an anomaly, a Diagnostic Data Set (DDS) is 
generated. The On Board Database Server (ODBS) stores both 
process and diagnostic data. All data are transmitted to the off-
board servers with their time-stamp every two hours. In case a 
proper transmission channel is not available, data are stored and 
transmitted when a connection is available. 

It is possible to implement some simple KMIs (mainly 
counters) directly in the TCMSs. 

2.2. The off-board infrastructure  
The off-board infrastructure consists of a powerful, 

redundant server that collects the data from all the ODBSs of 
the fleet. A software package, called Maintenance Software 
Package, makes data from the trains available to all the users.  

The maintenance staff should: 
- monitor the fleet; 
- detect and prevent faults and anomalies; 
- define the maintenance work schedule.  

The maintenance staff is coordinated with the various 
Regional Operation Rooms. 

3. MAINTENANCE POLICIES  
As explained in [10], with CBM an equipment is maintained 

only when it needs maintenance, so no unnecessary 
intervention is carried out. An effective CBM requires reliable 
data from the trains and powerful algorithms for their analysis. 
Effectiveness of maintenance evolves together with the 
evolution of the technological framework of the Company. For 
sake of simplicity, we split this evolution into four steps: 

- Reactive Maintenance: it includes cyclic interventions and 
overhauls during the stops at the depot. No train-ground data 
transfer is required; 

- Remote Maintenance: the maintenance staff can access the 
ODBS for monitoring the status of each vehicle. Alarms are 
received real-time; 

- Reactive Maintenance: the maintenance staff uses rules and 
algorithms for monitoring the trains and communicates with 
the on-board personnel in case of detection of anomalies. 
Maintenance schedule is dynamically updated; 

- Condition Based Maintenance: KMIs and algorithms are 
used for calculating the residual life of components. 

A cost/benefit analysis is necessary for deciding to switch 
from one step to a higher one. 

4. DATA ANALYSIS FOR CBM  
Techniques of Data Mining are useful for analysing the large 

amount of data from the trains stored in the centralized 
database. Data mining means to extract useful information 
hidden in the data and to present them in the more simple and 
usable ways. Data mining methods use statistical approach and 
different mathematical techniques suitable for managing data 
and database to look for correlations. 

Essentially, a database for CBM purposes contains in the 
rows different objects - in our case different vehicles of the 
same fleet - and in the columns the properties and the 
parameters describing the behaviour of the object (see Figure 
1). A typical locomotive has about 2000 operating parameters, 
split between about 10 major subsystems. A subset of these 
parameters may give useful information for maintenance 
purposes. The case histories at pos.5 show some examples of 
maintenance-related parameters for specific subsystems. 

Considering such a database structure, it is possible to define 
three possible approaches to the data analysis: 

- Horizontal: analyses the same property of different 

 
Figure 1. Example of CBM oriented database.  



 

ACTA IMEKO | www.imeko.org December 2016 | Volume 5 | Number 4 | 58 

vehicles; 
- Vertical: analyses different properties of the same vehicle; 
- Mixed: it is a mix of the previous approaches, i.e. different 

properties of different vehicles or of each single vehicle. 
Each analysis is useful for achieving specific results, as 

described in the following paragraphs. 
For each analysis, it is important to define the KMIs of 

interest. 

4.1. Horizontal analysis 
Horizontal analysis allows checking and comparing the 

behaviour of a single vehicle/apparatus/subsystem with the 
other elements of the fleet. Therefore, the analysis may identify 
the vehicles with abnormal behaviours in the fleet. 

KMIs for comparative horizontal analysis should highlight 
the overall status of the machine, like the total number of stops, 
the average consumption, etc. 

The first and easiest horizontal analysis is the calculation of 
average values (xavg) and variances (σ). Machines with 
parameters outside the average value of the fleet with a band of, 
for example, ±σ or ±2σ become candidates for a more accurate 
and precise analysis. Considering random failures, we can 
assume a Gaussian distribution of the samples. The level of 
confidence for setting the alarm threshold for each parameter is 
set according to the sample distribution.  

Considering the example in Figure 2 over 10 vehicles, it is 
apparent to understand that machine 4 and 9 show an abnormal 
behaviour. 

4.2. Vertical analysis  
Vertical analysis considers the behaviour of a 

vehicle/subsystem over the time. Thus, this analysis shows the 
upcoming of abnormal behaviour or degradation of a 
parameter. Different approaches exist for the vertical analysis: 

- Trend analysis: is the study of the behaviour of a KMI over 
the time that may identify a deviation from the expected trend 
(see Figure 3). The alarm threshold is set on the difference 
between the current value and the linear regression of the 
historical data; 

- Anomalies analysis: with a long-term monitoring, it may 
identify anomalies in the behaviour of a KMI, like sudden 
changes or abnormal operations; 

- Statistical analysis: it may identify abnormal deviations in 
the historical behaviour of a KMI; 

- Signature analysis of a device: the signature of a device is 
the set of values that significant parameters of a machine or a 
system have in normal operating conditions or during the 
execution of a normal operation. Deviations from the signature 
indicates abnormal operations. Figure 4 shows an example of 
signature for a pneumatic actuator using the profile of the 
output pressure during an open/close cycle. 

4.3. Mixed analysis  
The mixed analysis may consider different parameters of the 

same machine/system or also different parameters of different 
machines. The main scope of the mixed analysis is to identify 
correlations between different parameters or events. 

The basic techniques used for the mixed analysis are: 
- Regression: it is necessary to find a formula that expresses 

the relation between different selected variables; 
- Correlation: an index quantifies the correlation between 

two or more variable. Several indexes exist for measuring the 
degree of correlation of different variables. One of the most 
used index is the “Pearson correlation index”. The research of 
correlation indexes requires the comprehension of the physical 
relations between the variables (to avoid incongruous results). 

 
Figure 3. Example of trend analysis. 

 
Variance = 1 σ 

 
Variance = 2 σ 

Figure 2. Example of variance analysis. 

 
Figure 4. Example of signature for a pneumatic actuator. 

8

9

10

11

1 2 3 4 5 6 7 8 9 10

8

9

10

11

12

1 2 3 4 5 6 7 8 9 10



 

ACTA IMEKO | www.imeko.org December 2016 | Volume 5 | Number 4 | 59 

5. A CASE HISTORY: BOMBARDIER TRANSPORTATION ITALY 
5.1. A short history 

This paper describes the evolution of maintenance for two 
very popular locomotives: model E483 and model E186 that 
run in almost all the European countries. Both vehicles today 
have a technological structure that allows the implementation 
of Remote Maintenance. Process and maintenance data are 
collected on-board, and are fully accessible by the maintenance 
staff. Both locomotives started about ten years ago from a 
status of simple reactive maintenance, and the implementation 
of on-board and off-board infrastructures was therefore 
necessary to achieve Remote Maintenance.  

 Today all vehicles are equipped with a diagnostic controller 
that collects all the diagnostic data and sends them to an 
Ethernet gateway that communicates with the off-board system 
via GPRS. Similarly, the controller of each sub-system on board 
collects and transmits all the process data, including 
environmental data like position, speed, ambient temperature, 
etc.   

The control centre in Vado Ligure downloads these two 
databases through the GPRS modem and makes data available 
to maintenance staff through the web portal MyBTFleet. For 
safety reasons, national rules or laws make a maintenance stop 
for each vehicle mandatory at fixed schedule (e.g. in Italy every 
6 months). Through CBM techniques the maintenance 
operators can define the list of maintenance activities (not 
safety related) to schedule at a given stop. 

5.2. Actions towards Condition Based Maintenance  
For increasing the level from Remote Maintenance to CBM, 

it was necessary to implement a system for the graphical display 
of the diagnostic data stored in the database to give to the 
Maintenance Staff an immediate view of the evolution of every 
parameter. A specific tool for the intelligent data sorting and 
visualization was created with Matlab. Figure 6 shows an 

example of graphical output for digital variables. Maintenance 
engineers may use this tool for soting variables using: 

- a defined period of time, 
- the ID code of the variables, 
- the sub-system, 
- the status (e.g. show the variables that were “OFF” in the 

period from x to y).   
The tool can also calculate simple KMIs like: 
- counters (in a defined period of time),  
- conditioned counters (e.g. operating time with Variable X 

“ON” and Variable “Y” greater than Z), 
- statistics for each signal or group of signals (number of 

positive or negative transitions, time ON/OFF, etc.).  
More complex techniques of data mining are possible, but 

the size of the database suggests prudence. The monitored 
vehicles are about 150, and the considered parameters are about 
500 for each vehicle, with a sampling time of 1 second (data 
transmission is only by exceptions).    

The next three examples refer to the Bombardier fleet of 
vehicles, and represent: 

 
Figure 5. Example of correlation between different variables. 

 
Figure 6. Sample display of the visualization tool. 

  

  

 



 

ACTA IMEKO | www.imeko.org December 2016 | Volume 5 | Number 4 | 60 

- a horizontal analysis for the identification of CBM rules for 
the main circuit breakers, 

- a vertical analysis based on the signature for the air 
compressor on board,  

- a complex analysis of multiple measurements of the wheels 
consumption for determining their expected end-of-live.  

The manufacturers of the MCBs consider an average 
number of operation equal to 1000 after two year, and of 4000 
after 8 years. Figure 7 and Figure 8 show that the Main Circuit 
Breaker of all the locomotives have a much lower number of 
operations (both AC and DC).  

5.3. Rules for CBM: Main Circuit Breaker  
The Main Circuit Breaker (MCB) connects the locomotive 

to the supply line, and it is one of the most critical components. 
The MCB may be either DC (code “IR”) or AC (code “IP”). 
The MCB is an electro-mechanical equipment enduring both 
mechanical and electrical stresses. We used the historical 
database of the fleet for analysing all the failures of the MCBs 
looking for KMIs related with the failures. This phase of the 
study uses the knowledge of the equipment. In other words, we 
pre-selected a subset of data according to the knowledge we 
have of the construction and operation of circuit breakers to 
reduce the variables of the problem. The following KMIs 
proved to be useful:   

- KMI#1: counter of normal opening/closing operations of 
the MCB (with no current); 

- KMI#2: counter of MCB trips (the MCB opens the short 
circuit current); 

- Condition#1: train speed > 3 km/h (the train is not in a 
station); 

- Condition#2: status of the pantograph (open/close). 
We carried out a mixed analysis in search of correlations on 

the entire fleet of locomotors E186 for the year 2013 that is 55 
vehicles for about 11 months (a locomotive remains in the 
depot for about 1 month a year).  

The two KMIs are further split into two classes using 
Condition#1. When the pantograph is open (condition#2), the 
counter is not increased. 

The average value of these KMIs per month, together with 
their standard deviation was calculated to find out vehicles with 
abnormal values. Through the study of the maintenance record 
of these vehicles, we invented the following heuristic rule:  

IF (KMI#1 average per month > 3.5) 
AND (KMI#2 average per month > 3.5) 
THEN (maintenance is required) 
No vehicle with smaller KMIs experienced any failure of the 

circuit breaker. On the other hand, 33 out of 100 circuit 
breakers with higher KMIs did not experienced failures. The 
proposed heuristic rule gives the following results: 

- number of maintenance interventions per year: 
  . with remote maintenance 55 
  . with CBM rules  15  

of which 10 necessary  and 5 unnecessary.  
The application of the rule on the fleet of vehicles for one 

year of operation reduces the number of unnecessary 
interventions from 45 to 5 and does not cause any missing 
maintenance for the MCBs that really need it. 

5.4. Rules for CBM: Air Compressor  
The air compressor in a locomotive is an essential part of 

the braking system. The compressor has two output sections 
that supply the main braking circuit at a rated pressure of 10 
bar, and the primary circuit at 5 bar. Pressure losses happen 
when the brakes operate and for the normal leakages of the 
circuits. The compressor is controlled in on/off mode by a 
pressure switch with a hysteresis of 0.05 bar around the rated 
value. Typically, manufacturers of compressors for railway 
applications suggest a complete revision after 12,000 working 
hours (cyclic maintenance). 

To have a more accurate maintenance indicator the 
“signature” of the compressor was defined. When the train runs 
in normal coasting between two stations, the leakages of the 
circuits cause the pressure losses. The compressor compensates 
these losses and maintains the pressure in the range of 4.95-5.05 
bar. The horizontal analysis of data shows that the recharging 
time during the coasting phase for a new compressor lays 
between 60 s and 70 s.  Longer periods, in a reasonable 
percentage, should be an indicator of an overcoming problem 
in the compressor, mainly due to malfunctioning of the 
bearings. 

The following quantities are sampled every second: 
- Date and time 
- Speed 
- Pressure in the main pipes 
- Pressure in the principal pipes 
- Running kilometres 
Data were collected for a period of two months on the fleet 

of 186 locomotors in The Netherlands.  
We identified the “signature” of sound compressors and 

used it as a benchmark for finding out compressors with 
abnormal operation. The term “signature” refers to a set of 
measurement or quantities that identify the behaviour of a 
given component. 

The proposed KMIs for the compressors considers: 
- the overall average duration of a “on” period for the 

compressor, 
- the average duration of a “on” period for the compressor 

during coasting, 

 
Figure 7. Number of IP-MCB operations per year.  

 

 

Figure 8. Number of IR-MCB operations per year.  



 

ACTA IMEKO | www.imeko.org December 2016 | Volume 5 | Number 4 | 61 

- the load cycle of the compressor when the train is running 
(percentage of time “on” compared with the total running 
time). 

Figure 9 shows a typical profile of the air pressure during 
normal operation of a train. Pressure (blue line) increases when 
the compressor is running (red line), and it decreases when the 
compressor is off. Coasting is detected considering the train 
speed (light blue line); when it is constant the train is coasting, 
and pressure losses are only caused by the pipes leakages. 
Pressure variation during braking phases depends on the action 
of brakes, and are not related with the compressor status. 

When the three KMIs that compose the signature are listed 
in a table like in Table 1, some anomalies can be detected. 

  As Table 1 shows, compressors work during the coasting 
phase an average time of about 70 s. A longer time and an 
increase of this time over different periods may identify an 
upcoming problem of the compressor. On the other hand, 
Table 1 shows that the average value of the working percentage 
of the compressor is around 25 %. Again, values higher than 
the average may indicate that a failure or a malfunctioning of 
the compressor is approaching. 

For this analysis, we consider the average value and the 
standard deviation σ to identify the elements or vehicles that 
show an abnormal behaviour. 

Considering the normal distribution and the standard 
deviation, Table 1 points out two vehicles that have a very 
abnormal behaviour: locomotive 186113 and 186144. 

The main pipe recharging time, during the coasting phase, is 
a good indicator of the compressor performance. It is plain that 
the performances of E186113 and E184144 have a completely 
different behaviour from the other vehicles of the fleet, and an 
accurate check of these compressors during the next 
maintenance stop is scheduled. 

As all the compressor manufacturers indicate, another 
important parameter for maintenance purposes is the total 
working hours of the compressor. Typically, a complete 
overhaul is recommended after 12,000 working hours. 

Considering the average daily mileage and the effective 

percentage of working hours, it is possible to estimate the total 
worked hours for each compressor, and to evaluate when it will 
reach the target of 12,000 hours. 

Table 2 shows the estimated end-of-life calculated for the 
compressors under analysis. With the present usage rate, the 

  Table 1. Comparison of KMIs for different locomotors. 

 

Table 2. Estimated remaining life of the compressor. 

 

 
Figure 9. Typical profile of pressure and compressor operations. 

#Vehicle Avg Time ON Avg Time on coasting Avg % operation # Samples
186111 90.83 74.48 25.30 % 6
186112 122.91 73.98 30.05 % 4
186113 213.59 126.45 48.46 % 6
186114 73.81 64.47 18.58 % 4
186115 75.49 59.74 16.98 % 14
186116 68.27 54.39 16.13 % 4
186117 88.11 71.74 25.36 % 8
186118 75.79 60.09 22.77 % 8
186119 76.03 60.17 18.64 % 9
186120 77.69 64.45 22.42 % 10
186121 90.03 75.18 26.03 % 6
186122 89.81 67.35 26.89 % 3
186144 121.95 92.87 32.81 % 5
186236 76.05 63.22 21.81 % 5

# Vehicle Daily working hours Est. Remaining life [years]
186111 4.44 7.41
186112 4.54 7.25
186113 9.08 3.62
186114 3.38 9.72
186115 3.07 10.72
186116 1.90 17.27
186117 4.58 7.18
186118 3.79 8.68
186119 3.12 10.54
186120 4.23 7.77
186121 4.16 7.89
186122 4.30 7.64
186144 5.25 6.26
186236 3.39 9.70



 

ACTA IMEKO | www.imeko.org December 2016 | Volume 5 | Number 4 | 62 

compressor of vehicle 186113 will reach the target working 
hours in 3.5 years, so it has to be the first to be overhauled. 

5.5. Rules for CBM: Wheel Profile Monitoring  
When a locomotive enters/exits the depot, the Automatic 

Vehicle Inspection System (AVIS) collects data using laser, 
thermal and optical imaging technologies. Data are stored and 
analysed and, when necessary, AVIS generates alarms and 
automatic maintenance orders (see Figure 10). 

The two main parts of AVIS are: 
- VEMS (Vehicle Equipment Measuring System) that 

includes all the sensors, lasers, and cameras that generate the 
raw measurement data; 

- ORBIFLO: the Bombardier software platform that 
analyses the data for assessing the health status of the vehicle 
and for implementing rules for maintenance.  

Figure 11 shows VEMS, a modular system that contains the 
following modules:  

- axle end temperature monitoring system; it measures the 
dimensions and temperatures of axes; 

- brake pad monitoring system: it measures the thickness of 
every brake pad, calculates brake pad wear rate and 
predicts when replacement is due; 

- brake disc monitoring system: it measures the brake disk 
thickness, the disk profile and its maximum wear depth; 

- wheel profile monitoring system: it measures the wheel 
profile and assess condition in comparison to several key 
markers (flange height, thread hollow, etc.); 

- pantograph wear monitoring system: it checks the carbon 
strip profile, its maximum wear depth, and localized chip 
size; 

- wheel damage monitoring system (WDMS): it measures 
flat spots on wheels and determines when a wheel must 
be re-profiled by means of thresholds on the various 
measures; 

- visual image capture system: through laser scanning and 
optical imaging, it captures many data points of the train 
exterior, permitting verification of any deviation from the 

vehicle profile or previous vehicle condition. It is possible 
to measure the car height, coupler height, to detect 
missing or displaced elements, open equipment boxes, 
foreign bodies, etc. It also detects and assesses damper 
leakage conditions and vehicle contamination (oil leakage, 
impact damage, graffiti, etc.). 

The study that we carried out uses the data of the wheels of 
a British fleet composed of 302 trains measured and collected 
by AVIS. Measurements on each vehicle include 16 wheels; 8 
right wheels (RHS) and 8 left wheels (LHS). 

The analysis considers the following wheel parameters (see 
Figure 12): 

- diameter of the wheel; 
- flange height; 
- flange thickness; 
- difference between the right wheel’s diameter and the left 

one; 
- tread hollowness. 
To facilitate the data analysis, the data management system 

of Figure 13 was created. A server runs OSISoft PI analysis 
software composed by PI Interface, PI Data Archive. The users 
are client of the PI server, and run PI Processbook and PI 
Datalink for visualizing the data, together with Matlab for 
implementing the researched maintenance algorithms.  

The combined use of PI Processbook for the dynamical 
graphic representation of the variables of more wheels 
simultaneously (Figure 14), together with specific analysis 
developed in Matlab environment, made it possible to find out 
a formula for predicting the lifespan of the wheels: 

𝐿𝐿𝐿𝐿𝐿𝐿𝐿𝐿 =  𝑇𝑇 
365

∗
𝐷𝑚𝑚𝑚 −𝐷𝑚𝑚𝑚  −

�𝑇𝑇 ∗𝐷𝑑𝑑𝑑𝑑�
365
�

�𝑇𝑇 ∗𝐷𝑑𝑑𝑑𝑑�
365
� − 𝛥𝐷

.                    (1) 

With conventional Preventive Maintenance policy, wheels 

 
Figure 10. The AVIS Block Diagram. 
 

 
Figure 11.  Vehicle Equipment Measuring System (VEMS). 

 
Figure 12. Wheel profile. 

 
 

 
Figure 13.   Data Management System with OSISoft. 



 

ACTA IMEKO | www.imeko.org December 2016 | Volume 5 | Number 4 | 63 

are re-profiled every 200 or 300 days, according to the region 
the vehicle works in. With a policy of CBM, the analysis 
showed that it is possible to calculate the remaining useful 
lifespan of a wheel starting from the values collected by the 
WDMS. Interpolating the available values, it is possible to 
predict when a wheel will run out of tolerance (see Figure 15). 

This algorithm may reduce the maintenance interventions 
for the vehicle wheels of about 10 %. Its application just started 
few months ago, and sufficient data are not yet available for 
evaluating its effectiveness. 

6. CONCLUSIONS 
Maintenance is one of the higher costs for a Company that 

controls a fleet of vehicles. On the other hand, maintenance has 
an important effect on the Quality of Service of the same 
Company. An intelligent maintenance policy has positive effects 
in terms of both costs and QoS. In the paper we propose a 
roadmap that may lead a Company from a cyclic maintenance 
policy (either costly or ineffective) to Condition Based 
Maintenance (maintenance only when required). Increasing the 
effectiveness of the maintenance requires a technological 
infrastructure for on-board data collection and transmission to 
a ground maintenance centre where data are stored and 
analysed.     

Raw data from the vehicles are to be processed and grouped 
into Key Maintenance Indicators (KMI) that become the basic 
elements to use for maintenance-oriented algorithms. The 
identification of KMIs for a device or sub-system is based on 
two milestones: 

- the technical knowledge of the construction and operation 
of the device or sub-system to find-out a set of parameters that 
may be useful for maintenance purposes,  

- the analysis of the historical database for cross-correlating 
the candidate KMIs and the behaviour of the device or sub-
system. 

The paper shows three examples of this analysis: 
- the KMIs and the rule for deciding when the Main Circuit 

Breaker of locomotive has to be maintained; 
- the KMIs and the rule for understanding when the air 

compressor of the locomotive need an intervention. A rule for 
calculating the expected end of live is also presented; 

- a real-time data analysis structure for the calculation of the 
wheel consumption and the prediction of the necessity of their 
re-profiling. 

These rules are implemented on line, and are used for 
sending automatic maintenance warnings to the maintenance 
staff. 

REFERENCES 

[1] A. H. C. Tsang. “Strategic dimensions of maintenance 
management”, Journal of Quality in Maintenance Engineering, 
Vol. 8, No. 1, pp. 7-39, (2002) 

[2] W. Wang, A. H. Christer. “A survey of maintenance policies of 
deteriorating systems”, European Journal of Operational 
Research, Vol. 139, pp. 469-489, (2002) 

[3] M. Bevilacqua, M. Braglia. “The analytic hierarchy process 
applied to maintenance strategy selection”, Reliability 
Engineering & System Safety, Vol. 70, pp. 71-83, (2000) 

[4] X. Zhou, L. Xi, J. Lee. “Reliability-centered predictive 
maintenance scheduling for a continuously monitored system 
subject to degradation”, Reliability Engineering & System, Vol. 
92, No. 4, pp. 530-534, (2007) 

[5] L. Mann Jr., A. Saxena, G. M. Knapp. “Statistical-based or 
condition-based preventive maintenance”, Journal of Quality in 
Maintenance Engineering, Vol. 1, No. 1, pp. 46-59, (1995) 

[6] M. Bevilacqua, M. Braglia. “The analytic hierarchy process 
applied to maintenance strategy selection”, Reliability 
Engineering & System Safety, Vol. 70, pp. 71-83, (2000) 

[7] T.Farinha, I.Fonseca, A.Simoes, M.Barbosa, J.Viegas, “New ways 
for terology through predictive maintenance in an environmental 
prospective”, WSEA Transactions on Circuits and Systems, Issue 
7, Vol.7, July 2008, pp.630-647  

[8] R. Dekker, P. A. Scarf. “On the impact of optimization models 
in maintenance decision making: the state of the art”, Reliability 
Engineering & System Safety, Vol. 60, pp. 111-119, (1998) 

[9] M. Bohlin, M. Forsgren, A. Holst, B. Levin, M. Aronsson, R. 
Steinert, “Reducing vehicle maintenance using condition 
monitoring and dynamic planning”, Railway Condition 
Monitoring, 2008 4th IET International Conference on, pp. 1-6, 
(2008) 

[10] P. Pinceti, M. Caserza Magro, E. De Paola, P. Firpo, “A 
technological infrastructure for implementing a policy of 
Condition Based Maintenance for a fleet of railway vehicles”, 3rd 
Int. Conference on Circuits, Systems, Communications, 
Computers and Applications (CSCCA ’14), Florence (IT), 22-24 
November 2014 

[11] G. Del Gobbo, M. Giovannuzzi, M. Romairone, P. Masini, S. 
Rizzo, F. Romano, M. Romeo, “The Telediagnostica System for 
Trenitalia E464 and E405 Fleet”, Ingegneria Ferroviaria, Vol. 2, 
pp. 137-160, (2012) 

[12] I. Zolotova, T. Lojka, “Online data stream mining in distributed 
sensor network”, WSEA Transactions on Circuits and Systems, 
Vol.13, 2014, pp.412-421. 

 
 

 
Figure 14.  Top: Diameter, Flange Height and Flange Thickness trends of the 
same wheel; Bottom: Diameter trends of two different wheels. 

 
Figure 15.  Flange Thickness (red line), its linear regression (blue line) and its 
limits (green and pink lines). 


	CBM for a fleet of railway vehicles: infrastructure and algorithms