The ISO 13381-1 Standards failure prognostics process

through an example.
Diego Tobon-Meja, Kamal Medjaher, Noureddine Zerhouni

To cite this version:

Diego Tobon-Meja, Kamal Medjaher, Noureddine Zerhouni. The ISO 13381-1 Standards failure
prognostics process through an example.. IEEE Macau. IEEE Prognostics & System Health
Management Conference, PHM2010., Jan 2010, Macau, China. sur CD ROM, 11 p., 2010.
<hal-00459269>

HAL Id: hal-00459269

https://hal.archives-ouvertes.fr/hal-00459269
Submitted on 23 Feb 2010

HAL is a multi-disciplinary open access

archive for the deposit and dissemination of scientific research documents, whether they are published or not. The documents may come from
teaching and research institutions in France or
abroad, or from public or private research centers.

Larchive ouverte pluridisciplinaire HAL, est

destinee au depot et `a la diffusion de documents
scientifiques de niveau recherche, publies ou non,
emanant des etablissements denseignement et de
recherche francais ou etrangers, des laboratoires
publics ou prives.

The ISO 13381-1 Standard's Failure Prognostics Process Through an Example

Diego A. Tobon-Mejia, Kamal Medjaher and Noureddine Zerhouni
FEMTO-ST Institute, UMR CNRS 6174 - UFC / ENSMM / UTBM
Automatic Control and Micro-Mechatronic Systems Department
24, rue Alain Savary, 25000 Besanon, France
[email protected]
Abstract - Industrial failure prognostics can be considered
as the key process of any condition-based maintenance
solution. However, contrary to fault diagnostics which is a
mature research and industrial work, failure prognostics is
a new field for which few applications exist. In the last
decade, the interest for this activity has led to some open
and industrial standards where the main objective is to
provide users with a guidelines allowing them to perform
failure prognostics for a large class of industrial systems.
However, these standards, rightly, do not emphasize on any
particular example to illustrate their content. The present
paper aims at explaining the process of failure prognostics,
presented in the standard ISO 13381-1, through an
electromechanical example. The purpose is to help beginner
researchers in the field of industrial failure prognostics to
assimilate the main tasks of the process proposed by the
standard. The prognostics process is chosen because it
represents the key task among the rest of topics proposed
and published by the standard. Thus, the comprehension of
this part is important to develop prognostics methods and
algorithms based on the solid recommendations given by
the international organization for standardization.
I. INTRODUCTION
The need of reducing maintenance costs, improving the
availability and reliability, and reducing or eliminating human
and environmental accidents is nowadays a deep wish of any
industrial company. To reach these objectives, it is necessary to
adopt an appropriate maintenance strategy among the existing
ones. Indeed, maintenance can be curative or preventive. For
the last case, maintenance tasks are either planned in advance
for the systematic one or achieved depending on the real state
of the system, which is the case for the Condition-Based
Maintenance (CBM). In this maintenance policy [1, 2] the
system is monitored by a set of sensors which allow to track its
state and decide whether a maintenance action is needed or not.
To help performing this, failure prognostics seems to be the
convenient process that allows to estimate the Remaining
Useful Life (RUL) of the system and thus taking appropriate
decisions. Contrary to fault detection and isolation (FDI) or
fault diagnostics, for which many research and application
works are done, failure prognostics research activities are at
their infancy stage [1, 3]. However, this field is gaining more
and more interest from both industrial and research
communities. During the last decade, many definitions,
methods and tools are proposed in the field of failure
prognostics, but this is still an open research area [4]. The
recent abundance of literature regarding failure prognostics

may be a positive point, but it can also be a source of

confusion, especially for those who are beginning to work in
this domain. The role of any standard, open or not, is then to
provide users with some guidelines to help them to accomplish
their expected missions. Thus, in the open standard [5] one can
find some of the basic elements that should be fulfilled in order
to perform failure prognostics. In the same way, the standard
ISO 13381-1 [6] defines failure prognostics, details the steps of
the prognostics process, gives indications on the monitoring
system and on how to estimate the confidence interval
associated with the calculated RUL and proposes some
mathematical tools which can be used to model the
degradation. However, as for any standard, it does not focus on
a particular application, neither gives any illustrative example.
The present paper aims at explaining a part of what is
published in the standard ISO 13381-1, namely the failure
prognostics process. Each step of the process will be
progressively explained on an electromechanical example.
Section 2 of the paper briefly summarizes the main points
addressed in the standard. In section 3 the electromechanical
example is presented, followed by the definition of the failure
modes given in section 4. Section 5 is dedicated to the
explanation of the prognostics process by using the example
presented in the previous section. Finally, a conclusion is given
at the end of the paper.
II STANDARD ISO 13381-1
There exists in the literature some discussions about
prognostics in the framework of CBM open standards [5, 7-9]
proposed by the OpenO&M initiative [10] but in this paper
only the one of the ISO is addressed. The standard published
by the ISO deals with the main guidelines and aspects one has
to take into account in order to perform failure prognostics on
engineering systems. The standard gives the definition of
prognostics, specifies the required types of data to consider,
presents the concept and the process of prognostics along with
all necessary parameters, descriptors and influence factors to
consider in order to well estimate the remaining useful life and
the associated confidence value of a given system. Note that the
standard uses the terminology estimated time to failure (ETTF)
instead to RUL. The document also indicates a set of
mathematical models which can be used to model the
degradation phenomena.
Among the topics presented in the standard, three main
aspects can be pinpointed and deserve more attention, namely
the required data, the general monitoring process and the
prognostics process. In the following, only the failure

prognostics process is discussed and illustrated on a simple

academic example.
III SYSTEM DESCRIPTION
For the explanation of the generic prognostics process
proposed by the standard, a degradation phenomenon in an oil
pump is considered. Pumping systems were chosen because, in
one hand, they are one of the most critical mechanical systems
in the industry and, in the other hand they are frequently used in
the validation of prognostics models. For example, in [11] oil
pumps were used with different contamination levels to test and
to validate the data-driven approach the authors have
developed and which is based on Hidden Semi-Markov Models
(HSMM). In [12] the authors have used the condition and the
event data of centrifugal pumps for validation too. Their model
was based on a feed-forward neuronal network where training
targets were the asset survival probabilities estimated using a
variation of the Kaplan-Meier estimator and a degradationbased failure probability density function (pdf) estimator.
For the present paper, the considered example is shown in
figure 1. The system is composed of three main components:
one induction motor, one hydrostatic bearing and one
centrifugal pump (the function of each component and the
energy flow is explained below). The system is monitored by
different sensors coupled with an acquisition system used to
assess the actual condition of the system and to perform its
health prognostics.

Figure 1: Pumping System.

The main components of the pumping system of figure 1 and
their corresponding functions are explained hereafter.
Induction motor: this system converts the electric
energy into a kinetic energy by the interaction of magnetic
fields and current-carrying conductors. This device is
supplied by an alternating current, and can be separated in
two main parts: an outside stationary stator that has coils
supplied with alternating current to produce a rotating
magnetic field, and an inside rotor guided by bearings linked
to the output shaft to generate a mechanical torque by the
rotating field.

Hydrostatic bearing: solely supports the bearing's

loads on a thin layer of liquid or gas. Hydrostatic bearings
are frequently used in high load, high speed or high precision
applications where ordinary ball bearings have short life or
high noise and vibration. In our example this bearing is used
to guide the shaft toward the pump and to avoid the shaft
deflection.
Centrifugal pump: used to move liquids through a
piping system. The fluid enters the pump impeller along or
near the rotating axis (which is guided by bearings) and is
accelerated by the impeller, radial flowing outward into a
diffuser or volute chamber (casing), from where it exits into
the downstream piping system. Here, the pump performs the
expected or final activity "move a fluid at a constant flow
rate and pressure".
Sensors: used to perform the prognostics and to assess
the actual condition of the main components. Three types of
sensors are used in the monitoring process: accelerometers
which can capture the information contained in the vibration
profile, pressure sensors that can monitor with confidence
the refueling process in the pump and finally, an ohmmeter
to measure the resistance of the stator coils.
Table 1 presents the nomenclature of the used sensors and
the monitored components.
The pumping system works under its nominal specifications:
the flow rate, the speed and the pressure are assumed to be
constant values during the time (it is supposed that the system
is well controlled and without any external perturbation). The
sensors used in the monitoring process are taken as inputs of
suitable acquisition card which allows the data sampling at a
constant rate and uses anti-aliasing filters at a correct frequency.
IV PROGNOSTICS INPUTS
The first step in building a prognostics system, as published in
the ISO standard, is the identification of the set of failure
modes (FM), their influence factors on each other and the
detection measures (descriptors) that allow to track the
evolution of the degradation. The international standard IEC
60812 [13] has presented a procedure named "Procedure for
failure mode and effects analysis (FMECA)", which helps the
identification of all the failure modes for a specific system, by
the analysis of its subsystems and components. Also, the
FMECA method classifies the FMs using risk priority numbers
(RPN) that are calculated with three failure mode parameters:
occurrence (Occ), detection (Det) and severity (Sev). So, the
FMECA allows the definition of the appropriate detection
method and measures to be used in the diagnostics as well as in
the prognostics of the failure modes.
Concerning the pumping system, the results of the
corresponding FMECA analysis are shown in tables 2 and 3.
For this system, only eight FMs are considered and they are
defined as critical. So, the quantification with the RPN was not
necessary, because each FM can compromise the ended
function. The critical components are: the two bearings and the
stator in the electric motor, the oil in the hydrostatic bearing
and the two bearings, the impeller and the seal in the

centrifugal pump. To track the failure mode evolution, different

descriptors were defined. Some of these descriptors were tested
in the literature and have shown promising results. The failure
modes and their corresponding descriptors are defined in the
following paragraphs.
For the bearings failure, the descriptor to be used is
the measure of the "health probability". This descriptor is
Abbreviation
AccB1
AccB2
Ohm
AccHX
AccHY
AccBP1
AccBP2
AccPI
PS1
PS2

Sensor type
Accelerometer
Accelerometer
Ohmeter
Accelerometer
Accelerometer
Accelerometer
Accelerometer
Accelerometer
Pressure
Pressure

based on wavelet packet decomposition and Hidden Markov

Models (HMM) for tracking the severity of bearing faults. In
this case, the diagnostics system estimates the probability
that the current measures are generated by the normal
condition HMM model. So, any decrease of this probability
is considered as an indication of an incipient fault (see [14]
for more details).

Component
Bearing 1
Bearing 2
Stator
Shaft
Shaft
Bearing 1 pump
Bearing 1 pump
Impeller pump

Subsystem

System

AC Motor
Hydrostatic bearing
Pumping System
Centrifugal pump

Seal

Table 1: Sensors, abbreviations and associated monitored componen

For the stator in the electric motor, the descriptor used
is the resistance, it can be measured by a combination of
voltage and current measurements or simply by using an
ohmmeter (which is the case in this paper) [15].
The oil degradation in the hydrostatic bearing can be
tracked as suggested in [16] by monitoring the center
alignment of the shaft. This measure represents the thickness
of the oil film.
The cavitation in the pump which deteriorates the
impeller by the implosion of the bubbles can be tracked by
the acoustic emissions in the same way as proposed by
Escaler [17].
The leakage is monitored by measuring the differential
pressure between the flow before and after the seal and
assuming that the only component which can cause a leakage
is the seal.
As this is presented in the standard, the initiation criteria of a
failure mode must be defined by set of values which alerts the
user about a failure mode initiation. For the pumping system,
no numerical values were defined as this is considered as an
hypothetical example. Moreover, the FMECA analysis sheet
proposed for this example allows to record these values. The
alarm value is highlighted in yellow in tables 2 and 3.
In the same way, the standard specifies the definition of
other detection values for a failure mode, namely the alert and
the trip (shut-down) values. The alert value is higher than the
alarm value, but lower than trip value (see Eq. 1). The alert
value represents an advanced state of degradation and informs
the maintainer that a maintenance task should be defined to
assess the real condition and to avoid the break down of the
system. In the corresponding FMECA sheet, the space to

register this value is also considered and must be defined if this

example wants to be replaced by a practical one. The blue
column in the FMECA sheet is there to indicate this value. For
the alert value definition, several phenomena must be
considered as proposed in the standard:
the confidence level of the prognostics,
the future production requirements,
the lead times of spare parts,
the required maintenance planning,
the work required to rectify the faults,
and he trend extrapolation and projection.

alarm alert

trip

(1)

The trip, or the shut-down value, specifies the value that the
parameter/descriptor reaches when the component must be
fixed without any exception. Normally, this value is lower than
the real break value and can be understood as the security
factor in strength of materials. This value can be defined from
standards, manufacturers guidelines or by experience. In the
pumping system, the FMECA sheet contains also a column to
take into account these values, (see the red colored column in
tables 2 and 3).
V PROGNOSTICS PROCESS
In the same as the ISO, some researchers and industrials are
aware of the importance of prognostics in the CBM policy and
this pushed them to make some efforts in their standardization.
For example, the cooperation between scientists and industrials
has enabled the development of a standard platform to build a
CBM software [9], called "Open Systems Architecture for
Condition-based Maintenance". In this platform the

standardization of the prognostics process is discussed [7]. The

previous standards and the ISO standard, are consistent in
different points, especially in the fact that the prognostics is a
process which comes after the monitoring process. This is
because the monitoring process is used to estimate the actual
health state of the system and also to generate data which are
useful for the prognostics process.
The ISO standard defines the prognostics as a sequential
process (see figure 2) with four main steps:
1.
Pre-processing: at this step the system identifies all the
existing failure modes, their relations,
symptoms parameters - descriptors and determines the potential future
failure modes.
2.
Existing failure mode prognostics process: a study of
all existing failure modes is then performed, the severity and
the Estimated Time To Failure (ETTF) are calculated.
3.
Future failure mode prognostics process: the most
probable future modes, the influence factors between them
and the existing modes are estimated. Also, the ETTF for the
future failure modes is calculated.
4.
Post-action prognostics: in this step the prognostics
system proposes the maintenance actions to be done in order
to avoid, reduce or delay the failure mode effects. After this,
a new prognostics is made by taking into account the
suggested actions, and the confidence about the estimated
time to failure is calculated.

2.
The second step consists in the identification of the
influence factors between the existing failure modes. To
perform this operation the prognostics system can use the
"effects on" column in the FMECA analysis (see tables 2 and
3) to find the relations between the FMs. Once this operation
is performed for the pumping system, by using the diagnostic

Figure 3: Pumping System.

results obtained in the previous step and the FMECA
analysis, the prognostics system is able to define the
relations displayed with orange color in figure 4.

Figure 2: Pumping System.

In the following, the pumping system presented previously is
used to explain the main steps of the prognostics process.
Pre-processing
During this step the following operations are made:
1.
The first step deals with diagnostics to identify all the
failure modes which exist at a present time in the pumping
system. In our case a Hidden Markov Model can be made for
each failure mode listed in the FMECA sheet as proposed in
[18] to detect the existing failure modes. For the studied
pumping system the result shown in figure 3 can be obtained.
In this application example, the existing failures are:
Two growing outer race defects in the bearing 1 of the
motor and in the bearing 2 of the pump. Both failure
modes come from the past. They are in fact an evolved
version of past failure modes. For this reason the
diagnostics system places these failure modes in the past
line.
Two new failure modes that the diagnostics system has
identified at the present iteration time, by using the last
monitored data of the descriptors D4 and D8. These failure
modes are an oil degradation in the hydrostatic bearing
and a leakage in the seal of the pumping system.

Figure 4: Existing relations between the failure modes

3.
Once the existing failure modes and their relations are
identified, the system has to retrieve all the information
available about these failure modes. In the pumping system,
the FMECA analysis (table 2 and table 3) has different
detection levels (alarm, alert, trip). The prognostics system
must then retrieve this information to track the degradation.
For example, for the descriptor "D1" which is used

Table 2: Page 1 of the FMECA analysis

Table 3: Page 2 of the FMECA analysis

to track the outer race break, the system registers the

initiation criteria value (alarm), the alert value, the trip value
and the actual value as shown in figure 5. This operation
must be made for all the descriptors of all the existing failure
modes.

Figure 5: Existing failure mode values.

Figure 6: Probable failure mode values.

4.
The last step in the pre-processing process consists in
determining the potential future failure modes, their
initiation criteria and the failure definition set points. For
example, in the pumping system the old maintenance
histories data can be used to find the most probable future
failure modes. For this, the prognostics system can proceed
by similarity study and analysis (similar diagnosed failure
modes with their associated monitoring values) on the
previous recorded situations to find the potential future
failure modes. The result of this step is shown in the green
box of figure 6.
Existing failure modes prognostics process
During this step the following operations are made:
1. The first step of this process aims at identifying the
actual condition of the overall system, and the severity of the
existing failure modes. For this, the prognostics system will
classify the failure modes from the most severe to less severe.
To perform this quantification, the system measures the
distance between the actual descriptor value with its
respective trip value, and the smallest distance defines the
most sever failure mode. For example, in the pumping
system if the prognostics system displays in the same
window (using scales factors) the descriptors "D1" and "D4"
which have the trends showed in the figure 7, the system
concludes that the outer race break FM in the bearing 1 of
the AC motor (M_B1_1) is more severe at the present time
than the oil degradation FM in the hydrostatic bearing (HB).
This is because the distance between its current value and its
trip value is shorter than the descriptor "D4".
2. The next step is the time projection of the
parameters/descriptors into the future. For this purpose, the
standard presents projection and extrapolation as trending
tools. The basic difference between trend extrapolation and
trend projection is that projection requires the estimation of

Figure 7: Severity of existing failure modes.

extrapolation curve is fitted only to the existing data. The
projection is then more accurate than the extrapolation, if the
system disposes enough information or is accurately
modeled and captures all the failure modes behavior.
Interested readers can refer to the reviews about several
prognostics approaches and methods published in [1] and in
[19]. In the example of the pumping system, linear
regression models coupled with past information can be used
to project the descriptors as proposed in [20] or in [21]. For
example for the descriptors "D1" and "D4" the system will
trend a curve like the dotted lines shown in figure 8.
3. The subsequent step consists in a simple analysis of the
projection curves. The prognostics system estimates when
the projected parameter/descriptor reaches its respective trip
value (failure), and estimates the difference between the
actual time and the future failure time. This time difference
is the estimated time to failure "ETTF". In the case of the
pumping system, the prognostics model makes the

intersection between the projected values and the trip value

defined for all failure modes in the FMECA sheet and finds
the shortest "ETTF". For example, in figure 8 the
prognostics system projected the descriptors D1 and D4, and
it is possible to see that the descriptor "D4" reaches first its
trip value, whereas the descriptor "D1" does not reach its
corresponding trip value. Then, the system will keep in
memory the "ETTF HB" (green circle in figure 8) which is
associated with the failure mode "HB" (oil degradation).

Figure 9: Most probable failure mode estimation.

Figure 8: Parameters/descriptors projection and ETTF.

4. Finally, the prognostics process must assure that the
projection is under the specified confidence level. In the
pumping system, to calculate the confidence interval, it is
possible to use for example the Monte Carlo method to take
into account the variation of the parameters' values as in [22]
to estimate the ETTF of a suspension system.
Future failure modes prognostics process
1. The first operation in this step is the determination of
the most probable future failure modes. To perform this
operation the prognostics system can use the initiation
criteria values and the projections presented in the previous
subsection. The system will make the intersection between
the "alarm" values of the potential failure modes obtained in
the step 4 of subsection Pre-processing. Then the failure
modes which drop in this class will be retained as the most
probable future failure modes, whereas the others will be
removed for the current iteration. In the pumping system it is
possible to find a situation as presented in figure 9, where the
descriptor projection does not reach the initiation criteria of
the short-circuit in the stator (M_S). So, this failure mode is
then removed and only three future failure modes are
retained and handled, as shown in the red box of figure 9.

2. The next step is similar to the step 2 of subsection

Pre-processing where the influence factors were estimated.
Here the system estimates the influence factor between all
the failure modes (existing and future ones). For the
pumping system, knowing the existing and the future failure
modes calculated in the previous step, the system can use the
information indicated in the FMECA sheet ("effect on"
column) to determine the influence between all the failure
modes. Figure 10 shows the influence factors in the studied
pumping system, the orange arrows represent the influence
between the existing failure modes, the red arrows represent
the influence between future failure modes and the green
arrows identify the relations between existing and future
failure modes. Some arrows are bidirectional meaning that
degradation in one component may affect directly the other
and vice versa. This phenomenon is important and has to be
taken into account because a failure mode which evolves
faster and which is near the trip value will degrade the
related components too.
3. This step is similar to steps 2 to 4 presented in
subsection Existing failure modes prognostics process, in
where the system estimates the ETTF for the future failure
modes. To do this, the prognostics system will verify the
projections by considering the influence factors, the trip
values and the actual descriptor value.After this, the system
will estimate and retain the shortest future ETTF. In the
pumping system some descriptors of the probable future
failure mode obtained at step 1 of subsection Future failure
modes prognosis process may reach their trip values. For
example the descriptor "D7" shown in figure 11 and which
tracks the degradation in the impeller of the pump will reach
the failure state. So, an ETTF can be estimated and
corresponds to the shortest future ETTF which must be
stored in memory. For the case presented in figure 11 the
"ETTF P_I" will be retained (green circle in figure 11).

3. Finally, the prognostics system will present to the

maintainer the results, which are: the critical FMs and their
corresponding ETTFs, the maintenance actions, the
confidence values and the validity conditions. The standard
suggests many error sources which can affect the accuracy of
the calculated ETTFs and confidence values. For example,
confidence values can be completely different for each
descriptor because they depend on many parameters, like the
quality and the quantity of available information, the nature
of the database, the data acquisition and processing methods,
etc. For the pumping system example, the prognostics system
will present to the maintainer two messages, namely:

Figure 10: Influence factors between all the FMs.

Post-action prognostics
1. Here, the system will identify one or more maintenance
actions which may delay, stop or eliminate the progression of
the critical existing failure mode and prevent the initiation of
future failure modes. For the pumping system, these actions
are recommended in the FMECA sheet. From the previous
steps two critical failure modes with short ETTF were
identified, they are the "ETTF_HB" and the "ETTF P_I".
The prognostics system will then suggest to perform an oil
renewal and a pump inspection or an impeller replacement.
These actions are aimed at avoiding the failure modes and
eliminating their influence on the other failure modes as
shown in figure 12.

Figure 11: ETTF fo the future FMs.

2. At this step the system will re-estimate the existing and
the future failure modes by taking into account the proposed
maintenance actions. For the pumping system example, after
having done the proposed actions the prognostics system will
store all the failure modes (future and existing) in memory.
They will then be used in the next iteration as existing failure
modes and their presence will be validated or invalidated by
the diagnostics system.

a) An oil degradation is present in the hydrostatic bearing,

the ETTF is "ETTF_HB" with a confidence value of "Y"
%, an oil replacement is recommended. This prognosis is
valid under the assumption that the production plan is the
"Production" plan and the weather conditions will stay as
specified in the "Weather" file.
b) The actual conditions and the monitoring history advise
the possibility of cavitation which will degrade the
impeller; a pump inspection or an impeller replacement is
then recommended with an associated confidence value of
"R" %. This prognosis result is valid under the assumption
that the production plan is the "Production" plan and the
weather conditions will stay as specified in the "Weather"
file.
Conclusions
Many interesting topics are presented in the ISO 13381-1, such
as the condition monitoring flow chart detailing the prognostics
process, the prognostics basic concepts and the influence
factors. The prognostics process presented in the standard has
retained the attention because it can be considered as the key
process of any CBM solution. The generic process presented
by the standard may allow the users to understand, develop and
make failure prognostics if the appropiated means are used.
The example presented in this paper helps the new users
interested in industrial prognostics to understand in an easy
way the process suggested by the standard. Indeed, the
prognostics process as presented by the ISO is ambiguous in
many topics like for example the concept of existing failure
modes and the evolution of past failure modes, which are not
explained in detail. Thus, new users can have problems in
understanding the prognostics process.
The work presented in this paper aimed at explaining as well
as possible the prognostics process of the ISO 13381-1
standard through an example. However, the paper did not give
any numerical quantification of this ETTF or about the
corresponding confidence interval. This paper did not also use
any specific modeling or projection mathematical tool, because
an hypothetical example was used, but some references about
appropriated projections methods are given. The choice of
appropriate modeling and projection tools for RUL (or ETTF)
and confidence interval estimation depends in fact on the type

Figure 12: Maintenance actions and their effects on the FMs.

of the available data, the degradation mechanisms, the the
complexity of the system, the prognostics precision, and on
many other environmental factors. All these points are some of
challenges that deserve to be developed in the future works.
References
1. Andrew K.S. Jardine, Daming Lin, and Dragan Banjevic.
A review on machinery diagnostics and prognostics
implementing condition-based maintenance. Mechanical
Systems and Signal Processing, 20(7):1483 1510, 2006.
2. G. Vachtsevanos, F.L. Lewis, M. Roemer, A. Hess, and B.
Wu. Intelligent Fault Diagnosis and Prognosis for
Engineering Systems. Systems. New Jersey, Hoboken:
Wiley & Sons, 2006.
3. Venkat Venkatasubramanian. Prognostic and diagnostic
monitoring of complex systems for product lifecycle
management: Challenges and opportunities. Computers &
Chemical Engineering, 29(6):1253 1263, 2005.
4. Aiwina Heng, Sheng Zhang, Andy C.C. Tan, and Joseph
Mathew. Rotating machinery prognostics: State of the art,
challenges and opportunities. Mechanical Systems and
Signal Processing, 23(3):724 739, 2009.
5. M.G. Thurston. An open standard for web-based
condition-based
maintenance
systems.
In
AUTOTESTCON Proceedings, 2001. IEEE Systems
Readiness Technology Conference, pages 401 415, 2001.
6. ISO. Condition monitoring and diagnostics of machines prognostics - Part 1: General guidelines. Int. Standard
ISO13381-1, 2004.
7. M. Lebold and M. Thurston. Open standards for
conditionbased maintenance and prognostic systems. 5th
annual maintenance and reliability conference (marcon
2001). Gatlinburg, USA, 2001.
8. J.W. Sheppard, M.A. Kaufman, and T.J. Wilmering. IEEE
standards for prognostics and health management. In
AUTOTESTCON, 2008 IEEE, pages 97 103, 2008.
9. Applied Research Laboratory Penn State University, The
Boeing
Company,
and
Machinery
Information

10.
11.

12.

13.
14.

15.

16.
17.

18.

19.

Management Open Standards Alliance MIMOSA. Open

systems architecture for condition-based maintenance OSA
- CBM, 2006.
OpenO&M. http://www.openoandm.org/, 2009.
Ming Dong and David He. A segmental hidden semimarkov model (hsmm)-based diagnostics and prognostics
framework and methodology. Mechanical Systems and
Signal Processing, 21:2248 2266, 2006.
Aiwina Heng, Andy C.C. Tan, Joseph Mathew, Neil
Montgomery, Dragan Banjevic, and Andrew K.S. Jardine.
Intelligent conditionbased prediction of machinery
reliability. Mechanical Systems and Signal Processing,
23(5):1600 1614, 2009.
Analysis techniques for system reliability - procedure for
failure mode and effects analysis (FMEA), 2006.
Hasan Ocak, Kenneth A. Loparo, and Fred M. Discenzo.
Online tracking of bearing wear using wavelet packet
decomposition and probabilistic modeling: A method for
bearing prognostics. Journal of sound and vibration,
302:951 961, 2007.
Peter Vas. Parameter Estimation, Condition Monitoring,
and Diagnosis of Electrical Machines. Clarendon Press,
1993.
Agnieszka Muszynska. Rotordynamics. Taylor & Francis
Group, 2005.
Xavier Escaler, Mohamed Farhat, Philippe Ausoni, Eduard
Egusniza, and Francois Avellan. Cavitation monitoring of
hydroturbines: Test in a francis turbine model. In Sixth
international symposium on cavitation, 2006.
C. Kwan, X. Zhang, R. Xu, andL. Haynes. A novel
approach to fault diagnostics and prognostics. In
International Conference on Robotics & Automation,
pages 604 609, 2003.
H.C. Pusey and M.J. Roemer. An assessment of
turbomachinery condition monitoring and failure prognosis
technology. The Shock and Vibration Digest, 31:365 371,
1999.

20. Y.G. Li and P. Nilkitsaranont. Gas turbine performance

prognostic for condition-based maintenance. Applied
Energy, 86(10):2152 2161, 2009.
21. Tianyi Wang, Jianbo Yu, David Siegel, and Jay Lee. A
similarity-based prognostics approach for remaining useful
life estimation of engineered systems. In International
conference on prognostics and health management, 2008.
22. Jianhui Luo, Krishna R. Pattipati, Liu Qiao, and Shunsuke
Chigusa. Model-based prognostic techniques applied to a
suspension system. Transactions on Systems, Man, and
Cybernetics, 38:1156 1168, 2003.
Biography
Diego A. TOBON-MEJIA was born in Medellin, Colombia,
on February 16, 1985. He received the B.Sc. and S.M., both in
2008, from the National Engineering School in Metz (France)
and the EAFIT University (Colombia). He made a
specialization in his last school year in research and
development in Design, industrialisation and innovation
supported by the Paul Verlaine University, the ENSAM and the
ENIM, all in Metz (France).
In his last year of Master he developed a decision support
application based in fuzzy logic and the PERT method to assess
the failure mode hazard in a Francis turbine that works in a
hydroelectric power station in Colombia, to ensure the systems
availability and security. Actually he performs his doctoral
studies at Franche-Comt University, in Besanon (France)
sponsored by Alstom Transport where he his a research
engineer. He is engaged in research on rotating machinery
failure prognostic at the FEMTO-ST institute and Alstom
transport.
Diego A. TOBON-MEJIA, was honored in 2002 with the
Excellence scholarship from the EEPP de Medellin
(Medelln Public Enterprises) to perform his studies in
Colombia. In 2006 he was honored by the The French Ministry
of Foreign affairs with the Eiffel Excellence scholarship to
continue his studies in France.
Dr. Kamal MEDJAHER is an associate Professor at the
French high school of mechanics and micro-techniques in
Besanon since September 2006, where he teaches control and
fault diagnostics and prognostics. After an engineering degree
in electronic, he has got his MS in control and industrial
computing in 2002 at the Ecole Centrale de Lille and his PhD
in 2005 in the same field from the University of Lille 1. Before
joining his current position, Dr. Medjaher has mainly worked in
the field of Fault Detection and Isolation (FDI). He has
particularly used the bond graph formalism to derive
mathematical models of physical systems and to generate
Analytical Redundancy Relations (ARRs) and residuals, which
are then used to detect and to isolate possible faults on the
system.
Since September 2006, Dr. Medjaher leads research works in
the field of failure prognostics and uses artificial intelligent
tools, particularly Dynamic Bayesian Networks (DBNs). Dr.

Medjaher co-authored eight published journal papers and more

than ten refereed conference papers.
Professor Noureddine ZERHOUNI received his engineer
degree from National Engineers and Technicians School of
Algiers (ENITA) in 1985. After a short period in industry as
engineer, he received his Ph.D. Degree in Automatic Control
from the Grenoble National Polytechnic Institute in 1991. In
September 1991, he joined the National Engineering School of
Belfort (ENIB) as Associate Professor. At this time, his main
research activity was concerned with modelling, analysis and
control of manufacturing systems.
Since September 1999, Noureddine Zerhouni is Professor at
the national high school of mechanics and microtechniques of
Besanon. He founded and is responsible of the research group
"Design and maintenance of mechatronic systems" (COSMI) of
AS2M Department within FEMTO-ST Institute. His main
research activities are concerned with intelligent maintenance
systems and e-maintenance. Professor Noureddine Zerhouni
has been and is involved in various European and National
projects on intelligent maintenance systems like FP5 European
Integrated Project of ITEA program (Information Technology
for European Advancement) PROTEUS, NEMOSYS (Naval EMaintenance Oriented SYStem) with DCNS, and AMIMACFAME (Reliability Improvement of Embedded Machines) with
ALSTOM and CEGELEC.