Content uploaded by Marios Kefalas
Author content
All content in this area was uploaded by Marios Kefalas on Nov 14, 2019
Content may be subject to copyright.
A Review: Prognostics and Health Management in Automotive and
Aerospace
Van Duc Nguyena, Marios Kefalasa, Kaifeng Yanga, Asteris Apostolidisb, Markus Olhoferc, Steffen Limmerc, and Thomas
B¨
acka
aLIACS, Leiden University, Leiden, 2333 CA, The Netherlands
d.v.nguyen@liacs.leidenuniv.nl
m.kefalas@liacs.leidenuniv.nl
k.yang@liacs.leidenuniv.nl
t.h.w.baeck@liacs.leidenuniv.nl
bKLM Royal Dutch Airlines, 1181 GP Amstelveen The Netherlands
Asteris.Apostolidis@klm.com
cHonda Research Institute Europe GmbH, 63073 Offenbach am Main, Germany
Steffen.Limmer@honda-ri.de
markus.olhofer@honda-ri.de
ABSTRACT
Prognostics and Health Management (PHM) attracts increas-
ing interest of many researchers due to its potentially impor-
tant applications in diverse disciplines and industries. In gen-
eral, PHM systems use real-time and historical state infor-
mation of subsystems and components of the operating sys-
tems to provide actionable information, enabling intelligent
decision-making for improved performance, safety, reliabil-
ity, and maintainability. Every year, a substantial number
of papers in this area including theory and practical applica-
tions, appear in academic journals, conference proceedings
and technical reports. This paper aims to summarize and
review researches, developments and recent contributions in
PHM for automotive- and aerospace industries. It can also be
considered as the starting point for researchers and practition-
ers in general to assist them through PHM implementation
and help them to accomplish their work more easily.
1. INTRODUCTION
1.1. General introduction
At 11:03 Eastern Daylight Time (EDT), Southwest Airlines
Flight 1380 from New York to Dallas, was at about flight
level (FL) 320 (an altitude of approximately 32,000 feet) and
climbing when the left engine failed. As a result most of the
Duc Van Nguyen & Marios Kefalas et al. This is an open-access article
distributed under the terms of the Creative Commons Attribution 3.0 United
States License, which permits unrestricted use, distribution, and reproduction
in any medium, provided the original author and source are credited.
engine inlet and parts of the cowling broke off. Fragments
from the inlet and cowling struck the leading edge of the wing
and fuselage, causing a rapid depressurization. After investi-
gations, the reason was found to be failure of a single fan
blade, due to a fatigue crack (Accident: Southwest B737 near
Philadelphia on Apr 17th 2018, uncontained engine failure
takes out passenger window, n.d.). On 30th of September
2017 Air France Flight 66 from Paris to Los Angeles suffered
an uncontained engine failure and made an emergency land-
ing at Goose Bay Airport, Canada. Investigations indicated
that the engine’s fan hub had detached and dragged the air
inlet with it during the flight (Incident: France A388 over
Greenland on Sep 30th 2017, uncontained engine failure, fan
and engine inlet separated, n.d.). We all know that, unfortu-
nately, failure has never been completely prevented although
much money has been spent for equipment maintenance.
According to the annual reports of the Royal Dutch Airlines
(KLM)1, the maintenance costs from 2013 to 2017 are 669,
665, 934, 1009, and 994 million euro, respectively. These
correspond to about 11% to 18% of the operation costs. In
September 2003, the Commission of the European Commu-
nity reported that repair and maintenance accounts for 40%
of the total lifetime costs of vehicle ownership (Taie, Diab,
& ElHelw, 2012). The consequent costs due to the equipment
failure are high. DHL has estimated that an AOG (Aircraft On
Ground) due to technical reasons for an A380 Airbus, costs as
much as 925.000 euro per day2. In the worst cases, the con-
1https://www.klm.com/corporate/en/publications/2015 Annual Report.html
2Source: Airbus China
International Journal of Prognostics and Health Management, ISSN2153-2648, 2019 023 1
INT ERNATIO NAL JO UR NAL O F PROGNOSTICS AND HEALT H MANAGEMENT
sequent costs could not be fully evaluated if the equipment
failure led to a bad accident.
Prognostics and Health Management (PHM), therefore, has
emerged over recent years as an approach and methodology
that has a great impact in all industries. PHM is an engi-
neering discipline that aims at minimizing maintenance cost
by the assessment, prognosis, diagnosis, and health manage-
ment of engineered systems. With an increasing prevalence of
smart sensing and with more powerful computing, PHM has
been gaining popularity across a growing spectrum of indus-
try such as aerospace, smart manufacturing, transportation,
and power generation (Ekwaro-Osire, Stephen, Alemayehu,
Fisseha M, & Carlos Gonalves, Aparecido, 2017). Regard-
less of application, one common expectation of PHM is its
capability to translate raw data into actionable information to
facilitate maintenance decision making. Sometimes, PHM is
referred to as system health management (SHM), integrated
systems health management (ISHM), vehicle health manage-
ment system (VHMS) or engine health management (EHM).
In general, PHM provides for viewing overall health state
of machines or complex systems and assists in making cor-
rect decisions on machine or system maintenance. A ro-
bust PHM system should be able to detect incipient com-
ponent or system fault, perform failure diagnostics, failure
prognostics, and health management. Failure prognostics is
the heart of PHM. It refers specifically to the phase involved
with predicting future behavior and the system’s useful life-
time left in terms of current operating state and the schedul-
ing of required maintenance actions to maintain system health
(Vachtsevanos, Lewis, Roemer, Hess, & Wu, 2006). The use-
ful lifetime left is often called the ’Remaining Useful Life
(RUL)’. RUL is typically a random variable and unknown,
and as such it must be estimated from available sources of in-
formation such as the information obtained in condition and
health monitoring(Si, Wang, Hu, & Zhou, 2011). The main
implementation steps for PHM consist of; i) defining criti-
cal component(s), ii) appropriate sensor selection for condi-
tion monitoring, iii) prognostics feature evaluation under data
analysis and iv) prognostics methodology and tool evaluation
matrices(Atamuradov, Medjaher, Dersin, Lamoureux, & Zer-
houni, 2017).
PHM applications can be classified into two main categories
based on how the PHM is applied to the system or to the prod-
uct (Sutharssan, Stoyanov, Bailey, & Yin, 2015): i) real-time
PHM (sometimes referred as online PHM or on-board health
monitoring), ii) off-line PHM. Most of the safety critical and
mission critical applications require the real-time PHM. Usu-
ally modern aircrafts, automobiles and so on have substantial
on-board monitoring capability that is based on the use of
data from real-time sensors. For example, an electric car pro-
vides the range which can be achieved with the current bat-
tery state of charge based on the real-time monitoring of the
battery. Another example is the autonomous unmanned ve-
hicles, which have embedded real- time on-board PHM used
to re-plan the mission and reconfigure the controls based on
the health diagnostic and prognostic information. Such capa-
bility requires the evaluation of the current state of the health
and also a prediction of the future state of the component/
systems health (Tang et al., 2008; Sutharssan et al., 2015).
Approaches dealing with PHM are generally classified into
four categories: reliability based, model-based, data-driven
and hybrid. Each approach has its own advantages and draw-
backs. This topic will be discussed in more detail later on in
this review.
1.2. Existing Review Articles on PHM
There are a few review papers on PHM approaches and ap-
plications. Hereby, we list some examples by following the
order of appearance, from the oldest to the newest.
Jardine et al.(Jardine, Lin, & Banjevic, 2006) summarized
and reviewed research and developments in diagnostics and
prognostics of mechanical systems implementing Condition
Based Maintenance (CBM) with emphasis on models, al-
gorithms and technologies for data processing and mainte-
nance decision-making. Realizing the increasing trend of us-
ing multiple sensors in condition monitoring, the authors dis-
cussed different techniques for multiple sensor data fusion.
Wheeler et al.(Wheeler, Kurtoglu, & Poll, 2009) provided a
review over PHM user objectives and how they are related
to metrics commonly used in diagnostics and prognostics. In
this paper, authors identified critical gaps within the user ob-
jectives and the engineering development. A detailed survey
on the objectives of different users of health management sys-
tems was presented. These user objectives were then mapped
to the metrics typically encountered in the development and
testing of two main systems health management functions:
diagnosis and prognosis. They found that although the met-
rics associated with diagnostic and prognostic algorithm and
system performance positively impact the user community,
there were gaps within the diagnostic and prognostic metrics.
Si et al.(Si et al., 2011) reviewed the statistical data-driven
approaches for RUL estimation. Here authors reported the
up-to-date modeling developments for estimating the RUL.
The review was centered on statistical data driven approaches
which rely only on available past observed data and statistical
models. The approaches are classified into two broad types of
models, that is, models that rely on directly observed state in-
formation of the asset, and those do not. They systematically
reviewed the models and approaches reported in the literature
and finally highlighted future research challenges namely i)
development of a RUL estimation model based on very few
or no data situations. ii) data fusion where multi-dimensional
condition monitoring (CM) input data must be dealt with. Be-
2
INT ERNATIO NAL JO UR NAL O F PROGNOSTICS AND HEALT H MANAGEMENT
cause it imposes a severe challenge to threshold-based models
which are mostly established under a single threshold level.
iii) development of a model the influence of external envi-
ronmental variables. This is a complicated issue since those
variables will impact on the observed CM variables which
in turn will influence the RUL estimation. If it is not done
properly, overfitting can occur, which may reduce the robust-
ness of the developed estimation model. iv) development of a
model which can deal with multiple failure modes for a single
component.
Lee et al.(Lee et al., 2014) reviewed research on PHM design
for rotary machinery systems. This paper provides a review
of the PHM field, followed by an introduction of a system-
atic PHM design methodology for converting data to prog-
nostics information. This methodology includes procedures
for identifying critical components, as well as tools for select-
ing the most appropriate algorithms for specific applications.
Visualization tools are presented for displaying prognostics
information in an appropriate fashion for quick and accurate
decision making. Industrial case studies are included in this
paper to show how this methodology can help in the design
of an effective PHM system.
A specific software, which we will be referring to in the rest
of this study, is the Commercial Modular Aero-Propulsion
System Simulation (C-MAPSS) and its datasets. This is a
run-to-failure software and the datasets are generated from a
turbofan engine simulation model (Saxena & Goebel, 2008).
The dataset was first published by NASAs Prognostics Cen-
ter of Excellence (PCoE) in 2008. The original purpose of
generating this dataset was to use it in a data challenge com-
petition in PHM08 conference, where PHM researchers were
invited to develop prognostic methods as part of the compe-
tition. However, since then, these datasets have been widely
used by researchers around the world for developing prognos-
tic approaches and results in many publications. Nonetheless,
during the first six years, it was difficult for the users to suit-
ably compare their results due to the absence of performance
benchmarking results and common misunderstandings in in-
terpreting the relationships between these datasets. In 2014,
Ramasso et al. (Ramasso & Saxena, 2014) wrote a review
paper to summarize these approaches and analyzed them to
understand why some approaches worked better than others,
how did researchers use these datasets to compare their meth-
ods, and what were the difficulties faced, so necessary im-
provements can be made to these datasets to make them more
useful. The paper establishes public knowledge that helps
users in prognostic algorithm development and aids in fulfill-
ing the underlying intent of data repository to facilitate algo-
rithm benchmarking development.
Liu et al.(W. Liu et al., 2015) reviewed the structure healthy
condition monitoring and fault diagnosis methods in wind tur-
bines. In this paper, authors reviewed the structure of wind
turbines and analyzed the different components of wind tur-
bines in order to detect the faults that may happen. They
mainly reviewed fault diagnosis methods of wind turbines in
the last three years (up to 2015). Some research results on
diagnosing wind turbine components were analyzed, such as
time-frequency analysis methods, vibration based methods,
voltage and current based methods, etc. The advantages and
drawbacks of these methods were compared in detail in order
to find the most suitable methods. The main purpose of this
paper was to supply some information on structure healthy
condition monitoring and fault diagnosis in wind turbines for
related researchers.
Two independent groups reviewed the data-driven approach
and algorithms for PHM (Tsui, Chen, Zhou, Hai, & Wang,
2015; Sutharssan et al., 2015). Tsui et al. provided main
concepts and mathematical formulations that help readers to
quickly catch the key ideas and guidelines of each method.
They also showed three examples to illustrate the implemen-
tation of PHM. The first example was to identify fault di-
agnosis on gear crack development. The best classification
accuracy used weighted knearest neighbor method and was
near 100%, which was very beneficial for early warning of
potential gearbox malfunction. The second example was to
predict RUL of rotational bearings. The results showed that
the prediction based on the data-driven method was accept-
ably accurate, which provides very informative warnings on
the potential failures. The last example was to predict RUL of
Lithium-Ion batteries using Particle Filter. In the experiment,
batteries were tested with full charging and discharging cy-
cles, under the constant-current/constant voltage mode. The
results showed that the prediction was better and the probabil-
ity density function (PDF) of RUL was narrower at the later
stage of the batterys life.
In parallel, Sutharssan et al.(Sutharssan et al., 2015) aimed
at reviewing the structure, state-of-the-art, and classification
of algorithms and methods used to underpin different exist-
ing data-driven PHM approaches. This paper discussed dif-
ferent algorithms and mathematical models under different
data-driven PHM approaches. They showed that each ap-
proach and algorithm has its own advantages and disadvan-
tages depending on the application, availability of the histori-
cal data, system specific knowledge, programmability and so
on. PHM applications also have many different individual
processes such as noise reduction, anomaly detection, fault
isolation and monitoring, state estimation, lifetime prediction
and so on. They concluded that the selection of the approach
and algorithm for each process of a PHM application plays
a key role and is an important factor for the accuracy of the
overall PHM methodology.
Coble et al.(Coble, Ramuhalli, Bond, Hines, & Ipadhyaya,
2015) reviewed PHM applications in nuclear power plants.
Here authors highlighted the key research needs and techni-
3
INT ERNATIO NAL JO UR NAL O F PROGNOSTICS AND HEALT H MANAGEMENT
cal gaps that must be addressed in order to fully realize the
benefits of PHM in nuclear facilities. They also reviewed the
state of the art and the state of practice of prognostics and
health management for nuclear power systems. They con-
cluded that research, development, and deployment of PHM
for nuclear power systems have largely lagged behind other
safety-critical industries.
Jung et al.(Jung & Ismail, 2015) attempted to provide an
overview about PHM trend and direction of PHM in automo-
tive industry. However, authors failed to do so. They just sim-
ply listed some publications of PHM on the battery, engine,
antilock braking system and electric power steering system.
But they did not review and discuss anything about methods
and results.
Vogl et al.(Vogl, Weiss, & Helu, 2016) had a review of diag-
nostic and prognostic capabilities and best practices for man-
ufacturing. This paper reviews the challenges, needs, meth-
ods, and best practices for PHM within manufacturing sys-
tems. This includes PHM system development of numerous
areas highlighted by diagnostics, prognostics, dependability
analysis, data management, and business. They pointed out
that the challenges and needs that must be overcome for the
widespread realization of PHM within manufacturing. Based
on current capabilities, the critical challenges are real-time di-
agnostic and prognostic methods, standards for PHM system
evaluation, and the integration of data within user-friendly
PHM systems. Specifically, this PHM system be both reli-
able and flexible for use with multiple data sources.
Elattar et al.(Elattar, Elminir, & Riad, 2016) wrote a literature
review in prognostics in general. To the best of our knowl-
edge, this is the first comprehensive vision about prognostics
as a part of PHM in a single literature review paper. Au-
thors focused on reviewing prognostics benefits, approaches,
applications, and challenges. They gathered a lot of sparse in-
formation about prognostics and combined all of these infor-
mation together to present an integrated work that shows the
importance of prognostics and its influencing rule in PHM.
They also clarified how the maintenance strategies can shift
from fail and fix to predict and prevent based on the proactiv-
ity in prognostics and how prognostics is the main building
block in CBM. They discussed the prognostics approaches,
their advantages and disadvantages, and how to use the suit-
able technique according to the prognostics problem defini-
tion. They also presented a lot of prognostics applications
which have been already deployed or are just an experiment.
Finally, they addressed the more challenging aspects in prog-
nostics and how the research community is trying to resolve
these challenges. This paper can be considered as a starting
point for new prognostics researchers.
Atamuradov et al.(Atamuradov et al., 2017) had a review of
implementation and tools evaluation of PHM for maintenance
practitioners. Authors presented a general view of PHM and
its steps to provide prior knowledge for users, reviewed differ-
ent PHM approaches under model-based, data-driven and hy-
brid models, and discussed their merits and drawbacks. They
also reviewed previous and on-going research in bogie com-
ponents PHM to highlight problems faced in the railway in-
dustry. As a result of PHM literature review on bogie compo-
nents, they noticed that nearly all research conducted in bogie
health assessment is mostly limited to diagnostics rather than
prognostics tasks. Since railway vehicle bogies are critical
components, research on prognostics for asset health man-
agement is also crucial to provide a safe and comfortable ride
for customers.
The successful PHM applications in the industry require the
contributions from not only the field of reliability engineer-
ing and maintenance scheduling, but also the field of manu-
facturing engineering. In recent 20 years, production systems
of advanced manufacturing paradigms (e.g. mass customiza-
tion, reconfigurable manufacturing, sustainable manufactur-
ing and service-oriented manufacturing) have been developed
to exceed the traditional mass production paradigm. The rea-
sons that make system health management especially diffi-
cult include, individual machine deterioration, different sys-
tem structures, diverse production characteristics and expo-
nential scheduling complexity. To address these gaps, Xia et
al. (Xia et al., 2018) provided a review of the PHM work fo-
cusing on prognostics approaches for asset health, and main-
tenance policies for more ”informed” decisions. This paper
addresses recent advances in PHM for advanced manufac-
turing paradigms, to forecast health trends, avoid production
breakdowns, reduce maintenance cost and achieve rapid deci-
sion making. Furthermore, an in-depth look at future research
interests in this field is also provided.
Here we would like to introduce a comprehensive vision
about PHM with emphasizing the previous and on-going re-
searches in PHM for automotive- and aerospace industries.
This paper can also be considered as the starting point for
researchers and practitioners to assist them through PHM im-
plementation and help them to accomplish their duty more
easily.
The remainder of this review is organized as follows; section
II discusses failure modes and failure mechanisms, section
III diagnostics and prognostics, section IV introduces PHM
methods and section V covers the performance metrics. Sec-
tion VI and VII review PHM applications in the automotive
and aerospace industry, respectively and in section VIII we
conclude this review.
2. FAILURE MODE AND FAILURE MECHANISMS
An important part in PHM system is to clarify and identify
failure mode and failure mechanism. Failure mode is the
manner in which a system or component functionally fails. It
describes to what extent a certain function cannot be fulfilled
4
INT ERNATIO NAL JO UR NAL O F PROGNOSTICS AND HEALT H MANAGEMENT
anymore. Therefore, failure does not always imply the real
physical failure of a part, like fracture or melting, but could
also be the result of extensive deformation leading to rubbing
or seizure of a rotating part. Moreover, the definition of fail-
ure depends on what level is considered. Failure of a specific
part or subsystem does not automatically imply that the com-
plete system fails. For instance, a plant equipped with several
pumps does not stop when only one pump fails. In that case,
a failure occurs on the subsystem level (pump), but no failure
occurs on the system level (plant). In short, the failure modes
are generally related to the performance requirements of the
system (Tinga, 2013).
On the other hand, failure mechanisms are physical, chemi-
cal, thermodynamic or other processes that result in failure.
In PHM, knowing the failure mechanisms is a must to de-
velop model-based methods and is crucial to identify and
select features in the data-driven methods. Failure mecha-
nisms are categorized as either overload or wear-out mecha-
nisms. Overload failure arises because of a single load condi-
tion, which exceeds a fundamental strength property. Wear-
out failure arises as a result of cumulative damage related to
loads applied over an extended time (Pecht & Jie Gu, 2009).
Knowledge on these mechanisms, and especially the effect
of the governing loads on failure, are essential to understand
why, how and when components fail and how this can be pre-
vented. They are critically important in the PHM. The inter-
ested reader can find more details to failure mechanisms in
(Tinga, 2013). This book provides an overview of the most
important failure mechanisms. That includes static overload,
deformation, corrosion, fatigue, creep, wear, melting, thermal
degradation, electric failures, and radiative failures.
3. DIAGNOSTICS AND PROGNOSTICS
Diagnostics and prognostics are processes of assessment of a
system’s health. Diagnostics is an assessment about the cur-
rent (and past) health of a system based on observed symp-
toms. It deals with fault detection, isolation and identification
when a fault occurs. Fault isolation locates the fault to a spe-
cific component or area of a structure. Fault identification
determines the root cause of the fault. Often, these analyses
are completed in concert with each other; when an anomaly is
detected, the diagnostic system typically determines both the
location and cause of the fault given the available fault symp-
toms. Fault symptoms include the signatures that may help
diagnose the fault, including sensed data, features extracted
from sensed data, monitoring system residuals, and anomaly
detection results (Coble et al., 2015). Diagnostic capabili-
ties traditionally have been applied at or between the initial
detection of a system, component, or sub-component failure
and complete system catastrophic failure. In order to max-
imize the benefits of continued operational life of a system
or subsystem component, maintenance often will be delayed
until the early incipient fault progresses to a more severe state
but before an actual failure event. Practitioners reasoned that
if it were possible to use existing data and data sources to di-
agnose failed components, why would it not be possible to
detect and monitor the onset of failure, thus preventing fail-
ures before they actually hamper the ability of the operating
system to perform its functions. By doing this, mission relia-
bility would be increased greatly, maintenance actions would
be scheduled better to reduce system down time, and a dra-
matic decrease in life-cycle costs could be realized. More
recent diagnostic technologies enable the detection in much
earlier fault stages. The increase in this diagnostic capabil-
ity naturally has evolved into something more: the desire for
prognosis (Dong & He, 2007).
Prognostics is an assessment of the future health, it is a task
to determine whether a fault is impending and estimate how
soon and how likely a fault will occur. If an operator has the
will to continue to operate a system and/or component with
a known, detected incipient fault present, he or she will want
to ensure that this can be done safely and will want to know
how much useful lifetime remains at any point along this par-
ticular failure progression timeline. This is the specific do-
main of real predictive prognosis, being able to accurately
predict the RUL along a specific failure progression timeline
for a particular system or component. However, do not con-
fuse prognostic with RUL prediction. Because besides the
RUL prediction, a comprehensive prognostic should be able
to quickly and efficiently isolate the root cause of failures. In
this sense, if fault/ failure predictions can be made, the allo-
cation of replacement parts or refurbishment actions can be
scheduled in an optimal fashion to reduce the overall opera-
tional and maintenance logistic footprints. From the fault iso-
lation perspective, maximizing system availability and mini-
mizing downtime through more efficient troubleshooting ef-
forts is the primary objective (Vachtsevanos et al., 2006).
4. PHM A PP ROAC HE S
Commonly, prognostics approaches are classified into four
types (Elattar et al., 2016) namely i) reliability-based ap-
proaches, ii) model-based approaches, iii) data-driven ap-
proaches, and iv) hybrid approaches. Each approach has its
own merits and limitations. Nonetheless, generally speak-
ing, the complexity, cost, and accuracy of prognostics tech-
niques is usually inversely proportional to its applicability.
Increasing prognostics algorithm accuracy with low cost and
complexity is a big challenge. The prognostics system devel-
opers can benefit from this classification in the prognostics
approaches selection based on available data and their knowl-
edge about the engineer system. A key point about prognos-
tics approaches classification is to build a way to obtain a
standard methodology for prognostics applications develop-
ment within a standard framework.
5
INT ERNATIO NAL JO UR NAL O F PROGNOSTICS AND HEALT H MANAGEMENT
4.1. Reliability-based Approaches
Experienced-based prognostics, life usage model, or statis-
tical reliability-based approach are different names for the
same approach. These approaches are the simplest form
of fault prognostics as they require less detailed informa-
tion than other prognostic approaches. They are based on
the distribution of event records of a population of identical
items. Many traditional reliability approaches such as expo-
nential, Weibull, and log-normal distributions have been used
to model asset reliability. In practical applications, reliability-
based approaches can be implemented when historical repair
and failure data are available. These approaches do not con-
sider the failure indication (degradation) of an asset when pre-
dicting asset life (Gorjian, Ma, Mittinty, Yarlagadda, & Sun,
2010). In addition, these approaches are used mainly for un-
critical, unmonitored components that do not have a physical
model and are mass produced. We, therefore, exclude review-
ing applications of these approaches in this paper.
4.2. Model-based Approaches
The model-based method, sometimes referred as physics-
based method, is the most important approach in PHM be-
cause of its accuracy, precision, and real-time performance
(Elattar et al., 2016). It is a deterministic method and allows
the estimation and the prediction of the dynamic states. In
this approach, a physical/mathematical model for the system
or component is developed. This model is a real-time, math-
ematical representation of the system that is able to predict
the system degradation and failures. Additionally, it is able to
detect shifts from the nominal conditions when a simulation
based on the model runs in parallel to the actual process. To
establish this model, a thorough understanding of the physics
of the system/component is required and such a model’s re-
liability often decreases as the system complexity increases.
However, model-based methods do not require a large amount
of data and especially the data of the failure events. Be-
sides some physics-based models that are developed based
on physical principles/laws, the most common model-based
methods are Kalman filters (KF), extended Kalman filters
(EKF), unscrented Kalman filters (UKF), and particle filter
(PF).
Kalman filters were introduced as a fault isolation and as-
sessment technique for relative aircraft engine performance
diagnostics in the late 1970s and early 1980s (Simon, 2008).
More widely used by engineers and other physical scientists,
filtering problems are mathematical models for state estima-
tion. Kalman filters or linear quadratic estimation as they are
also known as, use measurements/observed values of a vari-
able of interest (the state variable) with the goal of making an
inference about it. They work in a two step process. Namely,
in the first step, the prediction step, the Kalman filter pro-
duces an estimate of the current state, along with its probabil-
ity distribution. Once the outcome of the next measurement
is observed, the previously produced estimates are updated.
It is a recursive procedure, which means that it only needs the
present observations and the previously calculated state and
its uncertainty matrix, to estimate the current state variable.
The latter hands them the advantage of running in-real time.
However, Kalman filters are linear model-based estimators,
which means that they assume linearity of the underlying dy-
namical system (Meinhold & Singpurwalla, 1983). In order
to overcome this and to address the non-linearities in either
the process model or the observation model or both, there ex-
ist the EKF and the UKF. The former assumes that the non-
linear functions are differentiable and linearizes about an esti-
mate of the current mean and covariance while the latter uses
deterministic sampling to form a new mean and covariance
estimate (Tahan, Tsoutsanis, Muhammad, & Abdul Karim,
2017) with a sampling technique known as the unscented
transform (UT) to determine a minimal set of sample points
(sigma points) around the mean.
The most popular model-based method is particle filters
(Chen Xiongzi, Yu Jinsong, Tang Diyin, & Wang Yingxun,
2011). PF method is a Sequential Monte Carlo (SMC)
technique for implementing a recursive Bayesian filter us-
ing Monte Carlo simulations. SMC methods are a set of
simulation-based techniques that provide an interesting ap-
proach to compute the posterior distributions of states. They
approximate the optimal filtering by representing the proba-
bility density function with a population of particles, which
are simply random samples (Daroogheh, Meskin, & Kho-
rasani, 2013). The basic idea is to develop a non-parametric
representation of the system state probability density function
in the form of a set of particles with associated importance
weights. The particles are sampled values from the unknown
state space and the weights are the corresponding discrete
probability masses. As the filter then iterates, the particles
are propagated according to the system state transition model,
while their weights are updated based upon the likelihoods of
the measurement given the particle values. They are a pow-
erful and effective tool for accomplishing state and parameter
estimation and allow for prediction in nonlinear dynamical
systems where the noise in the observations comes from an
arbitrary distribution and not just Gaussian. For more details
regarding PF, we refer the interested reader to (Arulampalam,
Maskell, Gordon, & Clapp, 2002).
4.3. Data-driven Approaches
As opposed to, a data-driven method is much easier to be
developed and applied in practical applications and is the
recommended technique when the feasibility study implies
a difficulty in obtaining a physical model. The low cost of
algorithm development and little knowledge required about
physics of the studied system, makes this approach preferable
6
INT ERNATIO NAL JO UR NAL O F PROGNOSTICS AND HEALT H MANAGEMENT
by prognostics system developers (Zhao, Liang, Wang, & Lu,
2017). In addition, data-driven methods provide a high reli-
ability and an effective computation, in spite of system com-
plexity. Data-driven approaches mainly rely on techniques
in the field of Artificial Intelligence (AI), which has many
ready-to-use tools that could be applied directly with minor
modifications. Nonetheless, compared to the physics-based
method, the data-driven method requires a large amount of
data, including both historical observation and current con-
dition monitoring data. In principle, the more failure events
are included in the data, the higher the accuracy of the esti-
mation obtained. However, failure events are normally rare
in any industry. In addition, this data is expensive and its ac-
cessibility is strictly limited for many reasons. Data-driven
approaches for PHM can be classified as falling within one
of the following two classes; i) statistical approach, and ii)
machine learning approach.
The statistical approach uses statistical parameters, such as
mean, variance, median and so on, to make predictions based
on known or unknown underlying probabilistic distributions.
Statistical approaches are generally considered to be simple
if the underlying statistical property (i.e. probability distri-
bution) is known. This type of approach is called parametric
approach. Statistical parameter estimation techniques and hy-
pothesis testing can be applied in this case to detect the pres-
ence of anomalies in the data. Here we list some examples of
the statistical approaches. These include; hypothesis testing,
analysis of variance (ANOVA), maximum-likelihood (ML)
estimation, Gaussian mixture modelling (GMM), Wilcoxon-
Mann-Whitney test, Bayesian networks (BN), hidden Markov
model (HMM), and principal component analysis (PCA).
However, machine learning approaches make predictions
based on acquired data (such as healthy and failure data) by
converting the gathered data into useful information which
can be used in conjunction with sensor data to provide fu-
ture predictions. Here we list some examples of the ma-
chine learning approaches. These include; nearest neighbour
(NN), neural networks, support vector machine (SVM), de-
cision tree, random forest, etc. Readers who are interested
more in the data-driven approaches can find more details in
review papers on the data-driven approach and algorithms for
PHM (Sikorska, Hodkiewicz, & Ma, 2011a; Tsui et al., 2015;
Sutharssan et al., 2015).
4.4. Hybrid Approaches
As previously mentioned, both model-based and data-driven
prognostics approaches have their own merits and limitations.
The hybrid (or fusion) prognostics approach, which is a newly
developing approach, aims to integrate the merits of these dif-
ferent approaches while minimizing limitations for better sys-
tem and/or component level health state estimation and RUL
prediction. It is a promising method because it can compen-
sate the lack of knowledge about the system’s physics and
the lack of data (Alia, Chebel-Morello, Saidi, Malinowski, &
Fnaiech, 2015; He, Williard, Chen, & Pecht, 2014a; Baraldi,
Compare, Sauco, & Zio, 2013). This fusion can be per-
formed either before the RUL estimation which is called pre-
estimate, or after the RUL estimation to obtain the final RUL
which is called post-estimate.
We, in this section, provided a brief overview about the PHM
approaches. Each approach has its own merits and limita-
tions. For practitioners, to select and implement a PHM ap-
proach is based on the application, the available data and their
knowledge about the monitored system. Case studies and ap-
plications of each approach will be reviewed separately for
automotive and aerospace industries.
5. PERFORMANCE METRICS
An important step in the successful deployment of a PHM
system is prognosis certification (Saxena et al., 2008). How-
ever, the community lacks on a standardized approach to
compare different methods in order for someone to iden-
tify the most suitable algorithm among a variety of possi-
ble choices. Additionally, there is an absence of a common
ground, that is, benchmark datasets or models on which the
techniques can be fairly compared. Performance metrics al-
low for the evaluation of different algorithms which can be
tested rigorously and evaluated by different measures before
they can be certified and thus employed in a real-world ap-
plication. Furthermore, the existence of metrics, allows for
establishing design requirements, specifications, guidelines
or characteristics that can be used consistently to ensure that
methods are fit for their purposes (Saxena et al., 2008), and
moreover is important for scientific, administrative and eco-
nomic reasons (Brier G.W. & Allen R.A., 1951). From a sci-
entific perspective, they matter due to the fact that they pro-
vide performance evaluations and therefore an objective way
to discern how prognostic models affect the quality of the pre-
diction. This thorough understanding yields valuable knowl-
edge and can guide research and development efforts in the
right direction. This refinement of the methods can result in
better performance scores justifying the investment in PHM
in areas that have not picked it up yet, as well as estimating
the return-on-investment (ROI).
5.1. Prognostics Metrics
Prognostics metrics can be classified into three broad cat-
egories, based on the end-use of prognostics information.
These are: algorithm performance metrics, computational
performance metrics and cost-benefit metrics. Since this re-
view paper is covering algorithmic methods in PHM, we will
briefly describe the two latter categories before moving on to
the former.
Cost-benefit metrics, such as life cycle cost, ROI, technical
7
INT ERNATIO NAL JO UR NAL O F PROGNOSTICS AND HEALT H MANAGEMENT
and total value, intend to measure the benefit gained by us-
ing prognostics. Table 3 in (Saxena et al., 2008) has a thor-
ough list of these metrics. Computational metrics on the other
hand, assess the computational performance of algorithms in
terms of time and memory space. These metrics are particu-
larly important in applications where there is need for a real-
time processing of data to make safety-critical decisions and
in embedded applications, such as on board computers of a
car or aircraft, which have limited space available.
Algorithmic performance metrics are usually based on ac-
curacy and precision, although algorithmic performance on
robustness and trajectory of the RUL also exists. Table 2
in (Saxena et al., 2008) has a thorough list of these met-
rics. Here we will discuss the ”new” metrics presented in
the same paper, as these specifically cater to PHM require-
ments. To keep the discussion concise we present three of
these. For a thorough reading we direct the interested reader
to (Saxena, Celaya, Saha, Saha, & Goebel, 2009) and to Table
4 in (Saxena et al., 2008).
5.1.1. Prognostic Horizon (PH)
An important question (requirement) to be asked when
performing RUL predictions for PHM is ”how far in advance
is it enough to predict with a desired confidence in the predic-
tions” (Saxena et al., 2009). The reason for this is of course
that it is desired to seek a prediction which is reliable but
also is enough time in advance before the actual end-of-life
(EOL), so there is time for appropriate maintenance action.
This leads to the Prognostic Horizon (PH) metric.
PH is defined as:
P H =EOL −i(1)
where:
i=min{j|(j∈`)∧(r∗−EOL·α≤rl(j)≤r∗+EOL·α)},
αis the allowable error bound around true EoL and thus iis
the first time index, when predictions satisfy α-bounds. Fur-
thermore, `is the set of time indexes when predictions are
being made, lis the lth unit under test (UUT), r∗is the ground
truth RUL, r(j)is the predicted RUL at time j. The PH is de-
clared as soon the corresponding predictions enter the band
of desired accuracy and its range resides in (tEOL −tp)and
max{0, tEOL −tE OP }, where EOP stands for end of predic-
tion. For instance an error bound of a= 1% identifies when a
given algorithm starts predictin estimates that fall within 1%
of the actual EOL. The more an algorithm predicts whithin
the desired accuracy scores the better its PH score is. As can
be seen in Fig. 1, PH1 is more desirable than PH2.
5.1.2. α−λPerformance
Another important requirement is determining whether the
prediction falls within specified limits at particular times,
that is how well an algorithm performs when additional data
Figure 1. Prognostic Horizon. Adapted from (Saxena et al.,
2010).
become available. Saxena et. al (Saxena et al., 2009) de-
fine α−λaccuracy, as the prediction accuracy to be within
α∗100% of the actual RUL at a specific time instance tλ.
In words, this metric determines whether a prediction falls
within 20% accuracy (α= 0.2) halfway to failure from the
time the first prediction is made (λ= 0.5). One needs to
evaluate whether the following condition is met:
(1 −α)∗r∗(t)≤rl(tλ)≤(1 + α)∗r∗(t)(2)
where αis the accuracy modifier, λis a time window modifier
such that tλ=tP+λ∗(EOL−tP)and tPis the time at which
the first prediction is made.
This metric is more rigid, compared to PH, as it requires pre-
dictions to stay within a cone of accuracy. See Fig. 2 for the
concept.
Figure 2. α−λaccuracy. Adapted from (Saxena et al., 2010).
8
INT ERNATIO NAL JO UR NAL O F PROGNOSTICS AND HEALT H MANAGEMENT
5.1.3. Relative Accuracy (RA)
Is similar to α−λaccuracy, but instead of finding out whether
the predictions fall within given accuracy levels, at a time
instants, it measures the accuracy levels. RA is defined as:
RAλ= 1 −|r∗(tλ)−rl(tλ)|
r∗(tλ)(3)
where tPis defined as before. See Fig. 3 for schematic rep-
resentation. An algorithm with higher relative accuracy is
desirable. In the figure we see two estimate curves (black and
red) and the ground truth RUL (blue). We see that at time in-
stant tλthe RA of the black estimate is (slightly) better than
that of the rest. It is also visible that the RA of the red esti-
mate decreases after time instant tλ, while that of the black
one increases.
Figure 3. Schematic representation of RA. Adapted from
(Saxena et al., 2009).
5.2. Uncertainties in Prognostics
To conclude this brief section, we must comment on uncer-
tainty representation and management (URM) as this is an
indispensable part of PHM. Accounting for uncertainties is
of paramount significance in prognostics. Uncertainties arise
from various sources such as: modeling uncertainties, mea-
surement uncertainties, operating environment uncertainties,
future load uncertainties, input data uncertainties. Such in-
formation is crucial for any prognostic estimate, otherwise it
is of limited use and cannot be incorporated in mission crit-
ical applications. The reason for this is that the single point
estimates that we described assume a deterministic algorithm
or additional reasoning. Due to all the sources of uncertainty
though, it is crucial that there is a confidence around the pre-
diction. There are numerous ways for this, like probability
distributions of the RUL instead of a single-point RUL esti-
mate. In (Saxena et al., 2010) and (Saxena et al., 2009), in a
very concise and detailed manner discuss the uncertainty is-
sues and propose solutions by modifying PHM metrics and
recommends suitable ways of graphically representing these
metrics.
To summarize, we briefly discussed the motivation and the
need behind performance metrics in PHM, by pointing out
the shortcomings and presenting certain proposed methods.
We described three of these methods, as we think they are
very representative and briefly discussed the need of incor-
porating uncertainty representation as uncertainty is inherent
in prognostics. Finally, it must be noted that the described
metrics are intended for offline evaluation of prognostics and
are not applicable for online cases. The reason for this is that
PHM performance evaluation is an acausal problem that re-
quires inputs from the events that are expected to take place
in the future. The reason is that one needs to know the true
EOL of the system to evaluate the prediction accuracy. On-
line evaluation will have to use methods to deal with uncer-
tainties associated with future operating conditions in partic-
ular (Saxena et al., 2009),(Saxena et al., 2010). This requires
future research in uncertainty representation.
So far we have covered PHM from a general perspective, in-
troducing its significance, goal, methods, and shortcomings.
In the remainder of this paper we will review PHM applica-
tions in the automobile and aerospace industries.
6. PHM IN THE AUTOMOTIVE INDUSTRY
According to a report in September 2003 published by the
Commission of the European Community, repair and mainte-
nance accounts for 40% of the total lifetime costs of vehicle
ownership (Taie et al., 2012). In 2010, Toyota recalled more
than 20 million vehicles due to technical issues, and nowa-
days software issues related to automotive controls account
for an increasingly large percentage of the overall vehicles
recalled. Therefore, a robust PHM system for automotive in-
dustry is required to overcome these issues. Recent advances
in sensor technology, remote communication and computa-
tional capabilities, and standardized hardware/software inter-
faces are creating a dramatic shift in the way the health of ve-
hicles is monitored and managed. Concomitantly, there is an
increasing trend towards the forecasting of system degrada-
tion through a prognostic process to fulfill the needs of cus-
tomers demanding high vehicle availability (Sankavaram et
al., 2009).
6.1. Classification of Automotive Sensors
Prognostics and health management generally combines
sensing and interpretation of environmental, operational, and
performance - related parameters to assess the health of a
product and predict RUL. Vehicles have very complex mecha-
tronic structures consisting of systems and subsystems. Nor-
mally any subsystem comprises electromechanical processes,
actuators, and sensors (Jeong et al., 2017). The sensors and
actuators are associated and controlled with an engine con-
trol unit (ECU) which manages and screens the procedure.
9
INT ERNATIO NAL JO UR NAL O F PROGNOSTICS AND HEALT H MANAGEMENT
Knowing the types, functions and applications of sensors is
required to develop a PHM system.
Due to rapid development of technology both the number
and type of sensors keep increasing. However, on the basic
level, the primary sensors in use today in automotive systems
are reviewed and classified according to their three major ar-
eas of automotive systems application. They are powertrain,
chassis, and body (Fleming, 2001). The powertrain encom-
passes every component that converts the engine’s power into
movement. This includes the engine, transmission, the drive-
shaft, differentials, axles; basically anything from the engine
through to the rotating wheels. Area of systems application
of the powertrain sensors are vehicle energy use, driveability,
and vehicle performance. Chassis, also known as a vehicle
frame, is the main supporting structure of a motor vehicle,
to which all other components are attached. The chassis is
considered to be the most significant component of an au-
tomobile. It is the most crucial element that gives strength
and stability to the vehicle under different conditions. The
main elements involved in chassis are steering, suspension
(tire, springs, shock absorbers and linkages), vehicle break-
ing and stability. Area of systems application of the chassis
sensors are mainly vehicle handling and safety. Anything that
is not powertrain or chassis is included as a body systems ap-
plication. It contains main elements such as occupant safety,
security, comfort, convenience and information. The main
sensors used in powertrain and chassis applications are listed
in Table 1 and Table 2, respectively.
Table 1. Sensors used in powertrain application
Functions Powertrain sensors
Cylinder Pressure, combustion-gas ion cur-
rent
Manifold Pressure, temperature
Turbo boost Pressure
Engine knock Vibration, combustion-gas ion cur-
rent
Air intake Mass flow, volume flow rate
Engine torque Magnetostrictive, cylinder-ciring-
induced, crankshaft speed modula-
tion
Camkshaft Rotational motion
Throttle, pedal Rotary motion
Fuel injection Pressure
Exhaust/catalyst Temperature, catalytic activity
Engine oil Pressure, level, quality (predic-
tive, ac-dielectric constant, cyclic
Voltammogram, thermal conductiv-
ity)
Coolant system Temperature, level
Fuel Tank/system Level, evaporation leak pressure,
flexible fuel composition
Transmission Gearshift position, input/output
shaft speeds, temperature, pressure,
torque.
Table 2. Sensors used in chassis application
Functions Chassis sensors
Brake System Pressure, fluid level
ABS anti-lock
braking
Wheel speed, pressure, lateral ac-
celeration
Brake-by-wire Pedal force/depression angle
Electric power
steering
steering wheel angle, steering
wheel torque
Vehicle stability Wheel speed, lateral acceleration,
yaw angular rate, steering wheel an-
gle
Active suspen-
sion
Strut displacement, chassis height,
body acceleration (vertical, lateral,
longitudinal), yaw angular rate, roll
angular rate, steering wheel angle
Tire pressure Wheel-to-wheel variance of rolling
speed, on-wheel sensor, wireless
Tire temperature On-wheel sensor, wireless
Figure 4. Important sensors of an automobile (Cheng et al.,
2010).
If we classify automotive sensors according to their func-
tions, we can do it as follows (Fleming, 2001): there are
six types of rotational motion sensors, four types of pres-
sure sensors, five types of position sensors, and three types
of temperature sensors, two types of mass air flow sensors,
five types of exhaust gas oxygen sensors, one type of engine
knock sensor, four types of linear acceleration sensors, four
types of angular-rate sensors, four types of occupant com-
fort/convenience sensors, two types of near-distance obstacle
detection sensors, four types of far-distance obstacle detec-
tion sensors, and ten types of emerging. Fig.4 shows im-
portant sensors and their positions on an automobile (Cheng,
Azarian, & Pecht, 2010).
10
INT ERNATIO NAL JO UR NAL O F PROGNOSTICS AND HEALT H MANAGEMENT
6.2. Automobile: From Diagnostics towards Prognostics
A diagnostic protocol normally communicates with Elec-
tronic Control Units (ECUs) and offers the application layer
services from reading out Diagnostic Trouble Codes (DTC).
The three well known diagnostic protocols are On-Board Di-
agnostics II (OBD-II), Unified Diagnostic Services (UDS),
and Remote Diagnosis and Maintenance Systems (RDMS)
(Taie et al., 2012). OBD-II refers to a vehicle’s self-
diagnostic and reporting capability. OBD systems give the
vehicle owner or a repair technician access to state of health
information for various vehicle sub-systems. UDS defines
the application layer, data link layer and the physical layer
of the diagnostic communication. It does not specify all de-
tails, but some are left out to the manufacturers. Remote
Diagnosis and Maintenance Systems (RDMS) is developed
thanks to recent advances in remote communications, innova-
tive mobile applications, human-machine interfaces, model-
based diagnostics, electronics and embedded system tech-
nologies. RDMS improves diagnostics methods and equip-
ment in order to accurately locate and diagnose any malfunc-
tions. Service technicians do not have to merely rely on vi-
sual and physical inspections alone to resolve vehicle prob-
lems. Moreover, these advances equip the automobiles with
the capability to share in-vehicle sensor and diagnostic in-
formation with remote computers, enabling vehicle diagno-
sis and maintenance be performed remotely while the vehi-
cle is being driven. They provide manufacturer specific re-
pair information according to the problems identified by the
Off/On-Board-Diagnosis systems. In addition, vehicle pa-
rameters can be monitored while the vehicle is being driven
to determine when maintenance is necessary.
There is a trend towards the forecasting of system degrada-
tion through a prognostic process to fulfill the needs of cus-
tomers demanding high vehicle availability. In 2012, Taie
and co-authors (Taie et al., 2012) presented a novel automo-
tive Remote Diagnosis Prognosis and Maintenance system
(RD&M). The elements of the proposed system include vehi-
cles, smart phones, maintenance service centers, vehicle man-
ufacturer, RD&M experts, RD&M service centers, logistics
carry centers, and emergency centers. The system promotes
the role of smart phones used to run prognosis and diagnosis
tools based on Least Squares Support Vector Machine (LS-
SVM) multiple classifiers. During the prognosis phase, the
smart phone stores the history of any forecasted failures and
sends them, only if any failure already occurred during the
diagnosis, to the RD&M service center. The latter will then
forward it to RD&M experts as a real failure data to improve
the training data used in prognosis classification and predica-
tion of the remaining useful life (RUL).
Classifying health status of the automatic gearbox was a
case study for this RD&M system. In this case study, the
training data was provided by the original equipment man-
ufacturer (OEM) system experts. Based on the relation be-
tween tachometer readings, vehicle speed readings and gear-
box temperature reading, the gearboxes are classified into
four classes such as ”OK”, ”RUL 40%”, ”RUL 10%” and
”NOK”. The gearbox is considered normal (OK) if the gear-
box temperature is normal and the gear ratio (ratio between
vehicle speed and motor speed) is within acceptable range, on
the other hand, failure (NOK) is detected if the gearbox tem-
perature was above normal regardless the values of the gear
ratio, finally there were two classes of warnings (RUL 40%
and RUL 10%) where the RUL was depending on the gear
ratio. The training was done on 100 examples of the above
mentioned three sensor readings. Cross validation was done
using leave one out technique to evaluate the classification
of LS-SVM versus the classical K-nearest neighbor K-NN.
The accuracies were 0.93 and 0.82 for LS-SVM and K-NN,
respectively.
Very recently, in 2017, Shafi and co-authors (Shafi, Safi,
Shahid, Ziauddin, & Saleem, 2018) developed a platform for
fault prediction of four main subsystems of vehicles: fuel sys-
tem, ignition system, exhaust system, and cooling system. It
is called ’Vehicle Remote Health Monitoring and Prognostic
Maintenance System (VMMS)’. In the VMMS, sensor data
is collected when the vehicle is on the move, both in faulty
condition (when any failure in specific system has occurred)
and in normal condition. The data is transmitted to the server
which analyzes the data. Interesting patterns are learned us-
ing four classifiers such as Decision Tree, Support Vector Ma-
chine, K-Nearest Neighbor, and Random Forest. These pat-
terns are later used to detect future failures in other vehicles
which show the similar behavior. The approach is developed
with the goal of expanding vehicle up-time and was demon-
strated on 70 vehicles of Toyota Corolla type. Accuracy com-
parison of all classifiers is performed on the basis of Receiver
Operating Characteristics (ROC) curves.
VMMS has three main layers including data generation, data
processing and remote monitoring. In the first layer, an OBD
II scanner is connected with the vehicle through OBD II port.
This scanner behaves like a bridge between vehicle and a
portable device, such as a mobile or a laptop which supports
Bluetooth. All the sensor’s data in form of Diagnostic Trou-
ble Codes (DTC) is generated when the vehicle is on the move
and sent to the portable device. In the data processing layer,
the first step is feature selection in which the data stream of
DTC is filtered in a feature selection process using the ex-
pert’s suggestions. Then a Principle Component Analysis
(PCA) is applied on the data set for feature reduction. After
that four classification algorithms are used in the classifica-
tion phase including Decision Tree, Random Forest, K-NN,
and SVM. Then, results are stored on the server for further
derivation which is used for fault prediction and remote mon-
itoring of the vehicle. In the remote monitoring layer, the
owner or concerned person of the vehicle can monitor the cur-
11
INT ERNATIO NAL JO UR NAL O F PROGNOSTICS AND HEALT H MANAGEMENT
rent condition of the vehicle remotely like fuel status, speed,
and current position. The driver or the owner of the vehicle
is notified about the failure of any subsystem of the vehicle
through automatic notification.
The VMMS was experimentally evaluated on the four main
systems of 70 vehicles of Toyota Corolla type including ig-
nition -, exhaust -, fuel -, and cooling. A stream of DTCs
was produced by sensors with a sampling frequency of 1 Hz
when the vehicle was on the move for each system under ex-
periment. Each reading is taken as an example and contains
around 20 DTCs. The dataset consists of 150 examples. DTC
generated by sensors is considered as an attribute or feature.
The feature value is set to 1 if that particular DTC is gener-
ated and set to 0 otherwise. The output of the system or class
label is also in binary form. If the system is in operation then
the output is set to 0 which means that the vehicle is in a safe
condition. If a fault occurred or the system breaks down, then
the output class label is set to 1. So the generated dataset is
completely binary in nature. That’s why the authors selected
Decision Tree, Random Forest, K-NN, and SVM algorithms
for classification as they perform well on binary data. All
classifiers are evaluated with 10-fold cross validation.
The performance of each algorithm on a particular subsystem
is evaluated on the basis of accuracy, precision, recall, and F1
score measures. Here accuracy is the percentage of the total
number of predictions that was found correct. Precision is
true positive accuracy. Recall is true positive rate. Lastly, the
F1 score is an accuracy indicator which is measured using the
precision and the recall. The precision, recall, and F1 score
are calculated as follows:
P recision =T P
(T P +F P )(4)
Recall =T P
(T P +F N )(5)
F1 = 2∗P recision ∗Recall
(P recision +Recall)(6)
where TP, FP, and FN are true positive, false positive and false
negative, respectively. The percentage accuracy of all algo-
rithms for all subsystems is shown in Table 3. In all cases,
results show that the performance of all algorithms is very
good and SVM is the best classifier. The lowest accuracy of
the SVM model is 96.6% which is achieved on the ignition-
and cooling systems, while the best accuracy is 98.5% which
is achieved on the fuel system.
RDMS and VMMS perform very well on the fault- detection
and classification. However, forecasting the RUL which is
the heart of any PHM system is still missing. There has not
been yet a robust PHM platform which can be used for prog-
nostics of the entirely vehicle. We are exploring our review
on prognostics for subsystems and components.
Table 3. Accuracy of the VMMS performance on the main
four subsystems
Classifiers Ignition Fuel Exhaust Cooling
DT 72.5 76.5 78.5 75.9
SVM 96.6 98.5 98.0 96.6
K-NN 81.9 94.6 89.9 94.6
RF 79.2 90.0 88.6 89.3
6.3. PHM for Battery
Batteries are a core component of many machines and are
critical to the system’s functional capabilities. Battery failure
could lead to reduced performance, operational impairment,
and even catastrophic failure, especially in aerospace and au-
tomobile systems (Goebel, Saha, Saxena, Celaya, & Christo-
phersen, 2008). Additionally, in terms of air pollution, green-
house gas emissions, and economy, using electric vehicles is
nowadays preferred by many people. Significant work has
been done to determine the states and conditions of batteries.
Readers who are particularly interested in this topic can find
more details in these review articles: Zhang et al. (J. Zhang
& Lee, 2011), Rezvanizaniani et al. (Rezvanizaniani, Liu,
Chen, & Lee, 2014), Berecibar (Berecibar et al., 2016), and
Lipu et al. (Lipu et al., 2018) and recent publications: You
et al. (You, Park, & Oh, 2016), Dang et al. (Dang et al.,
2016), Yang et al. (F. Yang, Xing, Wang, & Tsui, 2016), Ye
et al. (Ye, Guo, & Cao, 2017), Jafari et al. (Jafari, Khan,
& Gauchia, 2018), Tian et al. (Tian, Xiong, & Yu, 2019),
Razavi et al. (Razavi-Far, Chakrabarti, Saif, & Zio, 2019),
and Downey et al. (Downey, Lui, Hu, Laflamme, & Hu,
2019). In short, the main tasks of PHM for battery indus-
try are state-of-charge (SOC) estimation, current/voltage esti-
mation, capacity estimation and remaining-useful-life (RUL)
prediction. There exist many estimation methods including
model-based, data-driven and hybrid. Some examples of the
model-based methods are open-circuit voltage, current inte-
gral, internal resistance measurement, impedance measure-
ment, (discrete) Thevenin model, Coulomb counting, Parti-
cle filter, and (adaptive extended) Kalman filter, etc. Com-
monly used data-driven methods are ANN, SVM, RVM, Auto
- regressive moving average, and Fuzzy logic. Usually, the
model-based estimations have less computational cost and
high time efficiency. However, these methods perform well
only on absolutely clean and precise data. In reality, data can
contain uncertainties and noise. Data-driven methods per-
form better with such type of data. Nonetheless, these meth-
ods exhibit complex computation and need a large amount of
data for appropriate training.
Hybrid methods, therefore, have been developed to overcome
the limitations and thus improve the prediction performance.
12
INT ERNATIO NAL JO UR NAL O F PROGNOSTICS AND HEALT H MANAGEMENT
We here highlight some outstanding examples.
In 2012, Liu et al. (J. Liu, Wang, Ma, Yang, & Yang,
2012) developed a fusion prognostic framework to improve
the accuracy of system state long-horizon forecasting. This
framework strategically integrates the strengths of the data-
driven prognostic method and the model-based particle fil-
tering approach in system state prediction while alleviating
their limitations. In the proposed methodology, particle filter-
ing is used for system state estimation whereas a data-driven
method is used to predict future measurements for the model-
based method. The predicted measurements from the data-
driven method can be regarded as new measurements in the
model-based method when there is a lack of measurements
during long-term prediction. As an application example, the
developed fusion prognostic framework is employed to pre-
dict the RUL of lithium ion batteries through electrochemi-
cal impedance spectroscopy tests. The experimental results
demonstrate that the proposed fusion prognostic framework
is an effective forecasting tool that can integrate the strengths
of both the data-driven method and the particle filtering ap-
proach to achieve more accurate state forecasting.
In 2013, Xing et al. (Xing, Ma, Tsui, & Pecht, 2013) devel-
oped an ensemble model to characterize the capacity degra-
dation and predict the remaining useful performance (RUP)
of lithium-ion batteries. Their model fuses an empirical ex-
ponential and a polynomial regression model to track the bat-
tery’s degradation trend over its cycle life based on experi-
mental data analysis. Model parameters are adjusted online
using a particle filtering (PF) approach. Experiments were
conducted to compare the ensemble model’s prediction per-
formance with the individual results of the exponential and
polynomial models. The ensemble model demonstrated bet-
ter prediction performance (smaller prediction errors and a
narrower standard deviation). This is because this model bal-
anced the global and local regression performance. The de-
veloped model was evaluated on two different battery sets
with two different rated capacities. For both kinds of bat-
tery samples, credible and reliable prediction results were
achieved. However, there are some limitations in applying
this developed model. Firstly, temperature effect is not con-
sidered in model. Secondly, in some cases, it is difficult to
quantify the actual maximum capacity because the battery is
usually not fully discharged in every cycle. The authors sug-
gested to map the capacity of the partial discharge into the
equivalent fully discharged capacity before using the devel-
oped model. The transform relation can then be explored by
measuring the different voltages and finding the interaction
between the random cut-off discharge and fully discharged
voltage.
In the same year, Wang et al. (D. Wang, Miao, & Pecht,
2013) developed a capacity prognostic method to estimate
the RUL of lithium-ion batteries. This method consists of
a relevance vector machine and a conditional three-parameter
capacity degradation model. The aim of the relevance vector
machine is to find a few representative basis functions to de-
rive the prediction model by using sparse Bayesian learning.
The conditional three-parameter capacity degradation model
is used to fit the predictive values at the cycles of the relevance
vectors. Extrapolation of the conditional three-parameter ca-
pacity degradation model to a failure threshold is used to es-
timate the RUL of lithium-ion batteries. To illustrate how the
developed battery capacity prognostic method can be used,
three instance studies for batteries A1, A2 and A3 were con-
ducted. The results showed that the developed method was
able to predict the future health condition of lithium-ion bat-
teries. They found that as more capacity degradation data is
used to train the relevance vector machine, the accuracy of
the battery RUL prediction is improved.
In 2014, He et al. (He, Williard, Chen, & Pecht, 2014b) de-
veloped an artificial neural network-based battery model to
estimate the SOC, based on the measured current and voltage.
This model uses unscented Kalman filter (UKF) to reduce
the errors in the neural network-based SOC estimation. The
method was validated using LiFePO4battery data collected
from the Federal Driving Schedule (FDS)3and dynamical
stress testing. This UKF-based approach was implemented
to filter out the errors in the neural network estimation. They
reported the root mean squared (RMS) errors of the SOC esti-
mation were within 2.5% to 3.5% for different temperatures.
There are three main contributions of this study namely i)
a constructive searching approach was developed to find the
optimal neural network structure for SOC estimation, and ii)
a state-space model was developed that combines coulomb
counting and neural networks. Moreover, a UKF approach
was implemented to improve the neural network SOC esti-
mation under different temperatures. The developed method
eliminates the need to determine an open circuit voltage SOC
lookup table, unlike equivalent circuit model-based SOC es-
timation. The field collected data can be used to update the
neural network and increase the estimation accuracy. iii) This
method does not rely on the physics of batteries, since a neu-
ral network is a data-driven approach. As a result, the devel-
oped approach can be readily applied to batteries with differ-
ent chemistries.
In 2015, Zheng and Fang (Zheng & Fang, 2015) developed
a method that uses UKF with relevance vector regression
(RVR) to predict RUL of short-term capacity of batteries. A
RVR model is employed as a nonlinear time-series prediction
model to predict the UKF future residuals which otherwise
remain zero during the prediction period. The objective of
the integrated UKF-RVR is to predict the battery RUL in a
way that the battery model parameters can be continuously
and recursively updated by properly incorporating prediction
3https://www.epa.gov/emission-standards-reference-guide/epa-us06-or-
supplemental-federal-test-procedure-sftp
13
INT ERNATIO NAL JO UR NAL O F PROGNOSTICS AND HEALT H MANAGEMENT
information from the RVR method and the last true residual
value through model-based UKF method. The performance
of the proposed method was validated and compared to other
predictors with the experimental data of four batteries. Au-
thors claimed that, according to the experimental and analy-
sis results, the proposed approach exhibits high reliability and
prediction accuracy, which can be applied to battery monitor-
ing and prognostics, as well as be generalized to other prog-
nostic applications.
6.4. PHM for Suspension
Suspension is the system of tires, tire air, springs, shock ab-
sorbers and linkages that connects a vehicle to its wheels
and allows relative motion between the two. Suspension sys-
tems must support both road-holding/handling and ride qual-
ity, which are at odds with each other. The tuning of suspen-
sions involves finding the right compromise. It is important
for the suspension to keep the road wheel in contact with the
surface as much as possible, because all the road or ground
forces acting on the vehicle do so through the contact patches
of the tires. The suspension also protects the vehicle itself and
any cargo or luggage from damage and wear. The design of
front and rear suspension of a car may be different. In general,
suspension systems can be broadly classified into three sub-
groups: dependent, independent and semi-independent sus-
pensions (Dishant, 2017). These terms refer to the ability of
opposite wheels to move independently of each other. A de-
pendent suspension normally has a beam or driven live axle
that holds wheels parallel to each other and perpendicular to
the axle. This type of suspension system acts as a solid link
between two wheels such that any movement of one wheel
is translated to the other wheel. Also, the force is translated
from one wheel to the other. In contrast, an independent sus-
pension allows wheels to rise and fall on their own without af-
fecting the opposite wheel. This is a widely used suspension
system in passenger cars and luxury cars due to its advantages
over a dependent suspension system.
Springs are the main component in the suspension system
which help to reduce road shocks and vibration of a vehicle.
Depending on vehicles different types of springs are used and
they can be classified as: leaf spring, helical/coil spring, tor-
sion bar, rubber spring, or hydro-pneumatic spring. Springs
that are too hard or too soft cause the suspension to become
ineffective because they fail to properly isolate the vehicle
from the road. Vehicles that commonly experience suspen-
sion loads heavier than normal have heavy or hard springs
with a spring rate close to the upper limit for that the vehi-
cle’s weight. This allows the vehicle to perform properly un-
der a heavy load when control is limited by the inertia of the
load. Riding in an empty truck used for carrying loads can be
uncomfortable for passengers because of its high spring rate
relative to the weight of the vehicle. A race car would also be
described as having heavy springs and would also be uncom-
fortably bumpy. A luxury car, taxi, or passenger bus would
be described as having soft springs. Vehicles with worn out
or damaged springs ride lower to the ground, which reduces
the overall amount of compression available to the suspension
and increases the amount of body lean. Performance vehicles
can sometimes have spring rate requirements other than vehi-
cle weight and load.
Having a good maintenance scheduling for the suspension
system supports vehicle’s comfort and safety. Common
mechanisms that lead to suspension failure are crack propa-
gation, corrosions, chloride attack, creep, excessive deforma-
tion and deflection, damage accumulation, and fatigue dam-
age. Luo et al. (J. Luo, Pattipati, Qiao, & Chigusa, 2008)
and Jaoude et al. (Jaoude, 2015) focused on fatigue analy-
sis to predict the RUL of springs. They used real physical
principles/laws to establish their prediction models. These
physical principles include the stress-cycle curve, Rainflow
model (Matsuishi & Endo, 1968), Paris-Erdogan’s (Sobczyk
& Spencer, 1993) and Palmgren-Miner’s laws (Miner, 1945).
In these papers, a systematic model-based prognostic process
is presented to successfully predict the RUL of a system with
multiple operational modes, load conditions, environmental
conditions, and road conditions. However, the successful use
of the prediction models is limited to the simulation data. The
Application of the process to real-world suspension systems
is still missing.
In 2017, Yang et al. (C. Yang, Song, & Liu, 2017) used a
data-driven method to predict the RUL of hydro-pneumatic
springs. The main issues that cause failure in hydro-
pneumatic springs are gas leakage and oil leakage. The au-
thors developed a time domain fault feature method, based
on degraded pressure under the same displacement condition,
and a feature extraction method based on linear interpola-
tion methods and redefined time intervals. They then com-
bined this feature extraction method with a data-driven prog-
nostic method that was based on support vector regression
(SVR) to predict the failure probability and the RUL values
of these systems. Real vehicle historical data and simula-
tion data were used to verify the feasibility of the proposed
method. In both cases, they found a good agreement between
the predicted and the true values. However, the RUL could
be predicted ahead only a few hours due to limitations of
the available data. This could only help drivers to prevent
bad accidents, but it is not very meaningful for maintenance
scheduling.
In 2018, Luo et al.(H. Luo, Huang, & Zhou, 2018) developed
a health monitoring approach for a MacPherson strut suspen-
sion systems based on vibration signals. This approach can be
considered as a hybrid model, because it uses the Palmgren-
Miner rule and the Rainflow model to estimate the damage
and a neural network to predict the partial damage level. This
method consists of two major parts: multi-Gaussian fitting
14
INT ERNATIO NAL JO UR NAL O F PROGNOSTICS AND HEALT H MANAGEMENT
feature extraction and a long short-term memory (LSTM)
based damage identification. After the application of a multi-
Gaussian fitting method to extract meaningful and discrim-
inative features, a proper LSTM neural network is built to
predict the real-time partial damage level. Then the health
status, in the form of remaining useful life, is estimated.
The performance of the proposed health monitoring method
was experimentally verified by torsion beam suspension dura-
bility tests. To collect data sets, they conducted two major
experiments. The first experiment was to specify the driv-
ing cycles that can reflect real operating conditions. The sec-
ond one was to specify the measured signals. These aspects
are addressed as follows: under different driving cycles, the
suspension component bears different vibration load, which
causes obvious partial damage change. Therefore, the par-
tial damage level is mainly affected by driving cycles. To
make the monitoring system work effectively under real op-
eration conditions, driving cycles are selected according to
the guideline of comprehensiveness and distinctiveness. In
this work, bench and road tests under various driving cycles
were implemented to simulate the real operation conditions.
Bench tests were performed to obtain data sets for training
the LSTM model. The measured signals in the data sets of
road tests were used to validate the LSTM model. As it is of
high cost and time consuming to generate data by road tests,
they conducted bench tests to collect training data sets. For
the case of measured signals, the vibration signals are con-
sidered as candidate inputs for the health monitoring system.
This is because vibration is related to damage identification
of the suspension component and it is available on the Con-
troller Area Network (CAN) of a standard vehicle. For the
sake of applicability of the monitoring system, candidate in-
puts are composed of signals collected by common sensors:
the displacement and angular velocity of the vehicle body in
the x, y and z directions, the deformation of two springs and
two shock absorbers in the rear torsion beam suspension, the
deformation of the two springs in the MacPherson front sus-
pension and the vertical acceleration of four spindles in the
center of wheels. In both cases, they achieved striking pre-
diction accuracy, while requiring low computation time.
6.5. PHM for other Automotive Components
In 2016, Sankavaram and co-authors established an
inference-based prognostic framework for health manage-
ment of automotive components (Sankavaram et al., 2016).
The framework is called Cox-PHM. Cox-PHM uses data-
driven methods to detect fault diagnosis and degraded state
trajectories and to estimate the RUL of components. The
framework takes into account the cross-subsystem fault prop-
agation, a case prevalent in any networked and embedded
system. The key idea is to use a Cox proportional hazards
model to estimate the survival functions of error codes and
symptoms (probabilistic test outcomes/prognostic indicators)
from failure time data and static parameter data, and use them
to infer the survival functions of components via a so-called
soft dynamic multiple fault diagnosis algorithm. The average
RUL can be estimated from these component survival func-
tions. The proposed prognostic framework consists of two
phases: an offline training and validation (model learning)
phase, and an online testing (deployment) phase.
In the Cox-PHM, data is classified into three types namely i)
archived failure data (Type I data): age (or a surrogate func-
tion such as the mileage or operational time) of the vehicle
at the time of failure, i.e., age when an error code or symp-
tom is observed, or a component is replaced; ii) static envi-
ronmental and status parameter data (Type II data); and iii)
dynamic data (Type III data): time-series data and periodic
status data. The framework employs two key techniques: (i)
Cox proportional hazards model (Cox-PHM) (Klabfleisch &
Prentice, 2002), and (ii) soft dynamic multiple fault diagno-
sis (soft DMFD) inference algorithm (Singh, Kodali, & Pat-
tipati, 2009). The Cox-PHM computes the survival functions
of tests (or error codes), whereas the soft DMFD algorithm
is used to infer failing components in coupled systems. The
soft DMFD algorithm determines the most likely evolution
of component states that best explains the observed soft test
failure outcomes (i.e., complementary test survival probabili-
ties).
The training phase consists of two steps. In Step 1, Type I and
Type II data are used to compute static data-modulated sur-
vival functions for components, error codes, symptoms and
any observable test outcomes via the Cox proportional haz-
ards model. In the testing phase, when new feature data (Type
III dynamic data) is obtained via online data acquisition sys-
tems, the survival probabilities of error codes are estimated
using the Cox-PHM model as well as the baseline hazard
functions obtained from the offline module (from Type I and
Type II data). The RUL of a component at any time can be
computed from the survival function by defining a threshold
on the survival probability.
The framework was demonstrated on datasets derived from
two automotive systems: i) a dataset derived from an automo-
tive electronic throttle control (ETC) system simulator with
failure time data, static parameter data, and simulated test out-
comes; and ii) a dataset derived from an automotive regener-
ative braking system (RBS) with failure time data, and static
as well as dynamic parameter data obtained from simulation-
based fault injection experiments conducted in MATLAB /
Simulink.
In the first application, the prognostic approach is applied to
a dataset derived from an automotive ETC subsystem simu-
lator. The function of an ETC subsystem is to determine the
necessary throttle opening using sensors (such as the acceler-
ator pedal position, the engine RPM, and the vehicle speed)
and drive the actuator to obtain the required throttle posi-
15
INT ERNATIO NAL JO UR NAL O F PROGNOSTICS AND HEALT H MANAGEMENT
tion via a closed-loop control algorithm in the engine con-
trol module (ECM). The ECM also monitors the health of
the engine subsystem by processing parameter identifier data
(PIDs) collected from various sensors and generates diagnos-
tic trouble codes (DTCs or error codes) when a failure oc-
curs in any component. The dataset derived from the ETC
simulator consisted of 11 error codes (DTCs), 479 status pa-
rameters (PIDs) collected at the time of DTC firing, accel-
erated age of the vehicle and the repair/replacement actions
(i.e., repair codes (RCs)) performed on the system. A total
of five different repair codes are present in the training data.
That includes: RC1 (Accelerator Pedal Replacement), RC2
(Throttle Body Assembly Replacement), RC3 (Accelerator
Pedal Position Sensor Replacement), RC4 (Powertrain Con-
trol Module Replacement) and RC5 (Throttle Position Sensor
Replacement).
Under a single fault assumption, the authors showed the esti-
mated component failure probability for all repair codes. The
R2fits were quite high ( >90%) for the RC1, RC2, RC3 and
RC5 codes. For the RC4 code, although the R-square fit was
low (78%), at the long age axis, the estimated failure proba-
bility was higher than the actual component fault probability.
This suggests that the algorithm can estimate/predict the fail-
ing component before the actual component reaches the fail-
ing threshold - which is expected from an effective prognostic
algorithm. In addition to RC4, the failure probability of RC2
was also significant compared to other repair codes. This was
because RC2 was a hidden fault of RC4 and the algorithm
inferred RC2 as failing component with failure probability of
less than 0.5.
In the second application, the prognostic framework was ap-
plied to estimate the sensor and parameter degradations in a
regenerative braking system (RBS). The RBS consists of a
driver model, a powertrain controller, component controllers,
and the powertrain model. The driver model simulates the
drive cycles by setting accelerator and brake pedal positions
to achieve the desired vehicle speed. The output from this
block is the driver’s torque demand at the wheels; this acts as
the input to the powertrain controller (PTC). The PTC is the
supervisory controller making the high-level decisions that
affect the general state of the powertrain, the operating mode
of the vehicle, and accordingly deliver the torque requests to
the component controller. Subsequently, the component con-
troller converts these torque requests into component com-
mands. These commands are, in turn, treated as the actuator
commands by the individual components in the powertrain
model to achieve the requested torque and, consequently, re-
port the system status (e.g., engine speed, battery state of
charge) to the supervisory controller. The powertrain model
comprises of all the components that mimic the behavior of
hardware components, such as the engine, the battery, and the
motor.
There are 25 signals that were being monitored in the RBS
system. Important signals contained: i) sensor signals, such
as temperature, speed, and current measurements from the
hardware components in the powertrain model; ii) motor,
wheel, and engine torque demands sent from the powertrain
controller to the component controllers; and iii) component
commands sent from the individual ECUs to the hardware
components in the powertrain model. To demonstrate the
framework, two faults were considered such as motor speed
sensor fault and wheel inertia fault. These faults were injected
into the model as additive biases on the measured signals.
In both cases, the estimated component degradations were in
good agreement with the truth with an R-square fit of about
96%. An estimation of RUL at any time could be obtained
directly by defining a threshold on the failure probability.
To summarize, the authors presented a novel approach for
fault prognosis problem in coupled systems by combining
three types of data, i.e., failure time data, static environmen-
tal and status parameter data, and dynamic data. The frame-
work employed the Cox PHM to infer the survival functions
of components and subsequently estimated the component
degradations via the soft dynamic multiple fault diagnosis al-
gorithm. The framework are applied to two different auto-
motive cases to infer the component degradations (comple-
mentary survival functions) and the inference algorithm esti-
mated the component failure probabilities with a good R2fit.
Additionally, the authors claimed that although the proposed
framework was validated on automotive systems, it has the
potential to be applicable to a wide variety of systems, rang-
ing from aerospace systems to buildings to power grids. How-
ever, the framework does not include the uncertainty in RUL
estimation. It is very important for maintenance scheduling.
A recent hybrid framework developed by Nguyen et al.
(Nguyen, Limmer, Yang, Olhofer, & B¨
ack, 2019) provides
a method to generate RULs data with uncertainty of four es-
sential components of a passenger car, namely engine, brake
pads, springs, and tires. In this work, authors used CarMaker
simulation to simulate a fleet of 200 cars with different driv-
ing scenarios in New York City. They then used physics-
based and data-driven approaches to predict the RUL of the
above mentioned components. Their results show a good con-
sistency of both approaches. This framework can be used for
establishing an optimal maintenance schedule for a vehicle
fleet (Wang, Limmer, Olhofer, Emmerich, & B¨
ack, 2019),
such as a fleet of a taxi company.
For other components, several frameworks were developed
such as for engine (J. Wang, Mao, Zhu, Song, & Zhuo, 2009;
M.-H. Wang, Chao, Sung, & Huang, 2010; Beatrice, Guido,
Napolitano, Iorio, & Giacomo, 2011; Ko et al., 2014; Ashok,
Denis Ashok, & Ramesh Kumar, 2016, 2017), antilock brak-
ing (J. Luo, Namburu, Pattipati, Qiao, & Chigusa, 2010),
(axle-) gearbox (X. Zhang, Kang, Zhao, & Cao, 2013; Shao,
16
INT ERNATIO NAL JO UR NAL O F PROGNOSTICS AND HEALT H MANAGEMENT
Liang, Gu, Chen, & Ball, 2011; Saimurugan, Praveenku-
mar, Sabhrish, Sachin Menon, & Sanjiv, 2016), and steering
system (Ming Yu & Danwei Wang, 2014; Lin & Ghoneim,
2016). Nonetheless, in all cases, the success is limited to
diagnostic. The prognostic application, which is the most im-
portant part of PHM, is still either incomplete or missing. Es-
tablishing a PHM system for these components remains a big
challenge.
6.6. Discussions and Conclusions
Despite many PHM frameworks that have been developed
over the past decades, the success is still limited to detect-
ing and classifying faults. Although some works reported a
high accuracy in predicting the RUL, their focus mainly on
components that have a low complexity such as battery and
suspension. The reason is that, for these components, physic-
based models are possible to be composed. The prediction
accuracy, therefore, is commonly higher in comparison with
the data-driven method. For the other components and the
entire vehicle, building a robust PHM platform remains a big
challenge. The bottleneck is due to two main reasons: the
complexity of the (sub-)system and the available data. While
the high system’s complexity leads to many difficulties in de-
veloping model-based methods, the lack of data causes dif-
ficulties in applying data-driven approaches. To overcome
this bottleneck, researchers should focus on developing hy-
brid approaches. Moreover, open data sources should be
available to all researchers. For many reasons, data might
be accessible within particular companies and users, but it
is restricted for many other researchers. Making these data
sources available for all researchers will reduce the remaining
effort necessary for building a robust PHM platform. In addi-
tion, due to rapid development of sensor technology, remote
communication and computational capabilities, standardized
hardware/software interfaces, and smartphones, it should be
possible to develop a robust PHM app on the smartphone or
on the car dashboard for critical subsystems or components.
This app will be considered as a self-management tool for the
users, so they can actively manage the maintenance schedul-
ing for their vehicle by themselves in the most economic man-
ner.
7. PHM I N TH E AE ROSPACE INDUSTR Y
7.1. Introduction
Due to the high availability expectations from aircraft op-
erators and clients and the high costs incurred for mainte-
nance, when an aircraft is out of service (Vianna, Rodrigues,
& Yoneyama, 2015) or Aircraft On Ground (AOG), as well
as the supportability, testability and realibility of modern air-
craft (L. Yang, Wang, & Zhang, 2016), PHM systems play
a significant role in the aerospace industry, from which it
originated in the first place. Nowadays, it is very challeng-
ing for the industry now to keep its costs as low as possible
and to generate maximum revenue, since the last decade has
been turbulent for the aviation industry owing to the unprece-
dented rise in its commodities due to inflation (Paul et al.,
2008), as well as due to the fluctuation in the price of fuel.
Regarding the latter, IATA published that in 2017 the airline
industry’s estimated fuel bill reached 149 billion USD, more
than 3 times the figure of 2003 (estimated at 44 billion USD)
(IATA, 2018). The industry has to ensure that its asset utiliza-
tion is optimum and therefore, the maintenance management
system of the existing aircraft needs to be precise in order to
ensure that the aircraft spend maximum time in the air so as
to make the best use of its machinery. This is due to the fact
that maintenance is extremely expensive, mainly due to the
price of spare parts. As a result, one wants to maximise the
use and exploit the remaining life of the installed parts, keep-
ing them in operation by maintining and repairing them until
they exhaust their life limit and need to be replaced. This is
the role of PHM; to make sure that this happens and that no
part is exchanged prematurely. The notice of pending equip-
ment failure allows for sufficient lead-time so that necessary
personnel, equipment and spare parts can be organized and
deployed, thus minimizing both equipment downtime and re-
pair costs, and optimizing maintenance. Integration is one
of the trends of PHM systems, which means that PHM sys-
tems of engine and other aircraft parts are integrated with air-
craft PHM system (Shufen & Wanying, 2013). To the best
of our knowledge, however, there is no generic PHM frame-
work and architecture enabling communication and integra-
tion with the various contributing systems (R. Li, Verhagen,
& Curran, 2018), as well as no uniform design framework
of aviation PHM systems between countries (L. Yang et al.,
2016) and even between carriers/operators. In addition, a sys-
tematic method has yet to be established for developing and
deploying a PHM system, as the current ones are application
or equipment specific (Lee et al., 2014).
Among all the frameworks the most mature system is that
of the F35 aircraft, which constitutes the double-deck archi-
tecture. Using this multilayered framework, the system inte-
grates the airplane airborne information, and sends the nec-
essary information to the ground controls. This integrated
health management system determines the safety of the air-
craft and allows for the state management and maintenance
guarantee (S. Li, Zhang, & Wang, 2017). Another predictive
maintenance system, for a wide range of helicopters flown
by the military (rotocrafts), is called HUMS (Health and Us-
age Monitoring Systems), developed by UTC Aerospace Sys-
tems. This system can detect several different types of is-
sues using vibration analysis, ranging from shaft unbalance
to gear and bearing deterioration. In civil aviation the typical
representatives are the Airplane Health Management (AHM)
system of Boeing (L. Yang et al., 2016), the AIRcraft Main-
tenance Analysis (AIRMAN) system of Airbus and a more
17
INT ERNATIO NAL JO UR NAL O F PROGNOSTICS AND HEALT H MANAGEMENT
recent addition, namely, aircraft real-time health monitoring
system (AiRTHM) (L. Yang et al., 2016). For more detailed
information on these specific systems we refer the interested
reader to (Company, 2006) (Boeing) and (Holzer, 2011),
(Drappier, 2008), (Itier, 2007) (Airbus).
There is also a lack of standards for PHM system devel-
opment, data collection and analysis methods and data
management, although the PHM4SMS (Prognostics and
Health Management for Smart Manufacturing Systems)
of NIST (National Institute of Standards and Technology)
serves in designing such standards (Vogl, Weiss, & Donmez,
2014). Particularly in the aircraft industry, the published
standards for the guidance for PHM systems development
is MSG-3, developed by the Maintenance Steering Group
(MSG) of the Air Transport Association (ATA) and is
titled ”Operator / Manufacturer Scheduled Maintenance
Development”. It is used for developing maintenance plans
for aircraft, engines, and systems (Air Transport Association
of America, 2013) before the aircraft is in service and it also
helps in improving safety while at the same time reducing
unnecessary maintenance tasks (Vogl et al., 2014).
This chapter is intended to familiarize the reader with
the PHM systems in the aerospace industry, by introducing
concepts, presenting examples and discussing research
opportunities.
7.2. Classification of Sensors of the Gas Turbofan Engine
Here we briefly classify the most common and informative
measurements of a turbofan engine. An exhaustive list of sen-
sor measurements of the entire airframe and of the stations of
a turbofan engine is out of the scope of this review paper. The
authors decided to emphasize on the turbofan engine alone,
due to the fact that it is the core of the aircraft and one of the
most, if not the most, expensive asset of the airframe. Fur-
thermore, this is a starting point for researchers in the quest
for informative measurements. In the rest, we classify them
by type and by function.
In Table 4 we provide a classification of the most common
turbofan sensors based on their type and in Table 5 we present
a classification based on their application. We should note
here that N3 below, is not applicable in all engines.
7.3. PHM Methods in the Aerospace Industry
In this section we will give an overview of various PHM
methods used in the aerospace industry. To be more specific,
as stated in the introduction, CBM systems are founded upon
the ability to infer equipment condition using data collected
from sensors on monitored systems. In aerospace, these sys-
tems could be engines, thrust reversers, avionics, flight con-
trols, fly-by-wire, landing gear, braking, Environmental Con-
trol Systems (ECS), electrical systems and auxiliary power
Table 4. Turbofan sensors classified by type
Types Sensors
Temperature
Oil temperature, Total air/gas temperature
Static air/gas temperature,
Nacelle temperature,
Exhaust gas temperature (EGT)
Vibration
Core vibration, Fan vibration,
Core phase angle,
Fan phase angle
Pressure
Total air/gas pressure,
Static air/gas pressure,
Oil pressure
Spoll Speed Core speed (N2), Fan speed (N1), N3
Miscellaneous
Fuel flow, Oil quantity, Altitude,
Mach number,
Variable bleed valve (VBV) position,
Nacelle Anti-ice,
Wing Anti-ice,
Variable stator blades
(VSV) position
Table 5. Turbofan sensors classified by application
Functions Sensors
Gas Path
Total air/gas pressure, Static air/gas
pressure, Total air/gas temperature,
Static air/gas temperature
Engine Oil Oil temperature, Oil pressure, Oil
quantity
Engine Balance Core vibration, Fan vibration, Core
phase angle, Fan phase angle
Stalling/Surging VBV position, Wing anti-ice, Nacelle
anti-ice, VSV position
Thrust Setting
Engine pressure ratio (EPR), Fan
speed (N1), Core speed (N2), N3, Fuel
flow
Exhaust Exhaust gas temperature (EGT)
Flight Envelope Altitude, Mach number
units, to name a few. For each system there are also numerous
sensors, which reflect their components’ state and the overall
system health. For example, the current Airbus A350 model
has a total of around 6,000 sensors across the entire plane
and this number will increase as big-data analytics software
and broadband links become more affordable (Shih & Yang,
2014).
In the following sections we discuss prognostic and diagnos-
tic methods used in aviation as they are crucial for safety,
customer satisfaction and the airline revenue. We will em-
phasize more on prognostic applications in the industry, as
this type of predictive analytics is common across all fields
of industry, but is particularly valuable in commercial avia-
tion. In addition, as mentioned previously, diagnostics is in-
cluded in prognostics and thus, we can consider prognostics
as a natural extension of diagnostics. After all, ones needs
the latter to find the former (Sikorska, Hodkiewicz, & Ma,
2011b). Thus, we can consider the term prognostics to have
a broader definition and enclose activities such as supervise,
18
INT ERNATIO NAL JO UR NAL O F PROGNOSTICS AND HEALT H MANAGEMENT
monitor, detect and determine initial degradations, as well as
make fault/failure predictions.
Before we proceed it is important to disambiguate two terms,
which we will be seeing a lot below and are (oftenly) a source
of confusion. These are fault and failure. The former implies
that the system under observation is still operational, but can-
not continue operating without any maintenance action, oth-
erwise it will cease operating, resulting in a failure.
7.4. Applications of Prognostics Model-based Methods in
the Aerospace Industry.
Celaya et. al, (Celaya, Saha, & Wysocki, 2009) demonstrated
the feasibility of detecting failure precursors in semiconduc-
tor device behavior and using particle filtering (PF) which is
an intelligent prediction framework to derive RUL estimates.
Specifically they apply their method on the isolated gate bipo-
lar Transistor (IGBT), which forms the backbone of avionic
systems and plays a crucial role in vehicle controls, commu-
nications, navigation and radar systems. They underline the
importance of the estimation of an RUL probability density
function (PDF), rather than a single value given by the mean-
time-between-failure (MTBF) approach. In (Daroogheh, Me-
skin, & Khorasani, 2014), Daroogheh et al., extended the par-
ticle filtering scheme in order to predict the future behavior of
nonlinear dynamical system states and parameters based on
the observation forecasting concept with time series methods.
Specifically, they developed a fixed-lag dynamic linear model
with an adaptive length moving window for time series fore-
casting using a time-varying autoregression-moving-average
(ARMA) model with fixed and variable model orders. The
proposed method is applied for prognosis of gas turbine en-
gine health parameters for the next 60 steps ahead. In the
same context, (Bolander, Qiu, Eklund, Hindle, & Rosenfeld,
2009), Bolander et. al presented a model-based RUL predic-
tion for aircraft engine bearing prognosis, in which models
are updated by utilizing diagnostic data as a source of ad-
ditional knowledge in order to reduce the uncertainty in the
RUL prediction. The RUL prediction method is based on
a particle filter approach with Bayesian updating. Model-
based prognostics and diagnostics are effective even in the
absence of component characteristics. In this view, in a re-
cent paper by Zhang et al. (M. Y. Zhang, Liu, Hanachi, Yu,
& Yang, 2018), the authors showed the ability and effective-
ness of even a generic-developed model-based approach for
the degradation of the auxiliary power unit (APU), in the ab-
sence of component characteristics. Below we will review
this work in more detail.
In (M. Y. Zhang et al., 2018), Zhang et al. monitor the starter
degradation of APU, by designing, among others, a physics-
based model. An APU, is a small gas turbine which provides
pressurized air to start the main engines and electric power
for the aircraft before the main engines start. To initiate the
APU, an electric motor called the starter is used to provide
the power required for running the gas turbine. If the starter
is degraded, its mechanical power to accelerate the gas tur-
bine decreases and eventually the APU fails to start. This
can cause serious consequences such as AOG (Aircraft On
Ground), thus delays and potential safety hazards, since if
there is a loss of thrust in flight, the APU can be used to start
the main engines again.
Since the starter is connected to the gas turbine, it will af-
fect the performance of the APU if it gets degraded. Thus,
the APU gas-path measurements collected from the aircraft’s
sensors can be used to assess the state of the starter degrada-
tion. In this paper the authors select the Exhaust Gas Tem-
perature (EGT) as the relevant measurement for the starter
degradation, as it is measured by the majority of the control
systems (M. Y. Zhang et al., 2018). In this view, they con-
structed a physics-based model that estimates the EGT of the
APU based on thermodynamic principles, to detect anoma-
lies. To do this, they also modeled the two main anatomical
components of the APU, the compressor and the turbine. Due
to the lack of parameters and component characteristics, they
adopted a generic method to construct the models, which we
will present below.
For the development of the compressor model the authors
used the Buckingham πtheorem. They show that they can
determine the corrected mass in flow through a known rela-
tionship between the mass inflow of the compressor and the
compressor inlet temperature, pressure, shaft speed, flow co-
efficient and an empirically determined flow coefficient fac-
tor. In addition to showing how to calculate the compressure
pressure ratio, based on compressor work. Furthermore it is
shown how the polytropic and isentropic efficiencies can be
calculated. As a result they could determine the corrected
mass flow, pressure ratio and the isentropic (or adiavatic) ef-
ficiency. The latter is a parameter to measure the degree of
degradation of energy in steady-flow devices. Regarding the
turbine model, based on field expert knowledge that the mass
flow of the turbine relies mostly on the pressure ratio of the
turbine and little on the shaft speed, they could approximate
it as a function of the turbine pressure ratio. Similarly, the
turbine pressure ratio can be calculated based on the com-
pressor pressure ratio. Thus, with this, they could determine
the corrected mass flow, as well as the pressure ratio of the
turbine.
With the aforementioned measurements in hand along with
the flow compatibility and energy conservation principle, the
gas turbine transient model is finalized. Flow compatibility
assumes that mass flow through the turbine is equal to that
through the compressor. The EGT can be calculated thus
through the model, given as inputs the shaft speed at peak
EGT point Npeak, ambient temperature T01, and the altitude
that can be used to estimate the ambient pressure P01.
19
INT ERNATIO NAL JO UR NAL O F PROGNOSTICS AND HEALT H MANAGEMENT
This results in the following overall dependency:
EGTcal =EGTcal(T01 , P01, N , α, P Ws)(7)
where αis the angular acceleration and P Wsis the starter
output power.
For this study, data were collected from an Airbus A310 fleet,
including 14 starter nominal cases and 12 starter degraded
cases. During each starting cycle of the APUs, the authors
recorded four parameters. Namely, peak Exhaust Gas Tem-
perature EGTpeak and the aforementioned Npeak ,T01 and
the altitude. Based on these measured data and Eq. (1), the
model is constructed recursively as in Fig. 5.
Figure 5. Procedures for constructing the physics-based tran-
sient model. Adapted from (M. Y. Zhang et al., 2018).
The physics-based model was implemented to both nominal
cases and degraded cases. In figure 6 of the original paper,
we observe the EGT estimation based on the approach the
authors took, for the nominal condition.
There it can be seen that the physics-based model can effec-
tively estimate the EGT in a nominal case starter and although
there are differences between the measured data and the esti-
mated EGT values, the model has clearly predicted the vari-
ation trend of the EGT. At this point we should also take
into account that even though the authors dealt with a lack
of component characteristics, the results are acceptably accu-
rate, proving the effectiveness of model-based methods, even
in such a generic design.
Subsequently, the authors implemented the model in the de-
graded - starter cases. In figure 8 (a) and (b), of the original
paper, we can see the EGT estimations of the model and the
actual measured data in cases I and II. The models estimate
the EGT assuming the starter is in nominal condition, while
the measured data are collected with a degraded starter. In
both figures there are distinctive patterns, which indicate the
large deviations between the estimated values and the mea-
sured values. This can be seen for case 1 after and around
cycle 140 and for case 2 after and around cycle 370. The
reason behind this pattern is, as stated in (M. Y. Zhang et al.,
2018), more fuel is pumped in the APU to compensate for the
diminished turbine acceleration, due to the degraded starter.
This increased fuel flow results in higher EGT values, com-
pared to the nominal condition, under the same shaft speed.
This increasing deviation is an indicator of the incremental
deterioration of the starter and gives us an idea of what we
mentioned earlier. Namely, residuals are large in the pres-
ence of actual problems and malfunctions and small when
there are normal disturbances, noise and perhaps modelling
errors.
The previous example serves to justify in a comprehensive
manner the power of a model-based approach in PHM. In
many situations however, the complexity of the systems under
observation makes it impossible to grasp, even in a generic
way, this complexity and design robust and accurate mod-
els which can be used for prognostic purposes (Dragomir,
Gouriveau, Dragomir, Minca, & Zerhouni, 2009). Nonethe-
less, it is often the case that there exists historical data, which
capture the behaviour of measured signals or extracted fea-
tures from the incipient fault stage to even equipment failure.
In these cases, data-driven methods should and can be uti-
lized. In the next section we will discuss these approaches.
7.5. Applications of Prognostics Data-driven Methods in
the Aerospace Industry
Recently in (M. Y. Zhang et al., 2018), Zhang et. al, designed
a back-propagation, feedforward neural network to assess the
starter degradation of the APU using its gas-path measure-
ments. Feedforward NNs are the simplest form of artificial
neural networks where information moves in only one di-
rection from input nodes to output nodes. In a recent pa-
per by Ma et al. (Ma, Lu, Zerhouni, & Cheng, 2018) the
authors proposed an effective deep learning method, termed
stacked denoising autoencoder (SDA), for health state clas-
sification of aircraft engines considering the environmental
20
INT ERNATIO NAL JO UR NAL O F PROGNOSTICS AND HEALT H MANAGEMENT
noise. SDA proved to be effective in terms of cognitive com-
puting and pattern classification theory. Furthermore, the
proposed method beats its rivals, in terms of feature extrac-
tion due to the benefits of its deep architecture with a data
destruction process that is effective for robust feature repre-
sentation, where high-order features and shared representa-
tions can be learnt from the input samples by unsupervised
self-learning. The feasibility of the proposed method was
demonstrated using the 2008 PHM challenge datasets (see
(Ramasso & Saxena, 2014)). Continuing in this domain,
Zhong et. al (Zhong, Li, Lin, & Zhang, 2018) designed a
gated recurrent neural network (GRU Network) to predict the
EGT of a turbofan aero-engine. This method could address
the time series and nonlinear characteristics simultaneously
by the GRU blocks. The proposed algorithm was compared
to five other single prognostic methods, namely, an artificial
NN (ANN), support vector regression (SVR), extreme learn-
ing machine (ELM) and ensemble prognostic methods ran-
dom forests-based ELM (RF-ELM) and average aggregation
ELM (Avg-ELM). The proposed method achieved the best
prediction accuracy and acceptable prediction stability. Re-
current NNs, developed in the 80s, are a class of NNs that
capture time dynamics. They constitute a superset of the tra-
ditional feedforward NNs that can process information across
time steps. Specifically, due to their internal state (memory)
they can process a sequence of inputs, granting them the abil-
ity to model temporal dependencies and are thus suited for
tasks in which input and/or output consist of sequences of
points that are not independent. For a more thorough under-
standing of RNNs we urge the interested reader to (Lipton,
2015). Also in (Vatani, Khorasani, & Meskin, 2015), Vatani
et. al predicted the degradation trends of a gas turbine engine
by studying their effects on sensored data (i.e. temperature)
by using an RNN as a first approach, as well as a nonlinear
autoregressive model with exogenous input (NARX) neural
network architecture. In (Zou, Ma, Fang, & Yi, 2011), Ke-
Xu et al. designed a particle-swarm optimized NN for space-
craft prognostics. In (X. Li, Ding, & Sun, 2018), Li et. al
use a deep convolution NN (DCNN) for estimating the RUL
and they demonstrate the effectiveness of their method using
the C-MAPSS dataset for aero-engine unit prognostics. Con-
volutional neural networks (CNN) are a class of deep, feed-
forward NNs mostly used in analysis of visual imagery, that
exploit the local dependencies of visual information (Lipton,
2015). For a more thorough understanding of CNNs and
their mathematical formulation, we direct the reader to (Wu,
2017).
Neural networks allow the investigation of complex systems
without the need for any knowledge or assumption about sys-
tem structure. They are sophisticated modelling techniques
capable of modelling problems that are analytically and in-
herently difficult and for which conventional approaches are
not practical, including complex physical processes with non-
linear, high-order, and time-varying dynamics (Ahmadzadeh
& Lundberg, 2014). NNs however (at least standalone) do
not take into account uncertainty bounds arising from differ-
ent sources like process noise, measurement noise and an in-
accurate process model.
In contrast to NNs, relevance vector machines (RVM) and
Gaussian process regression (GPR) take into account the
width of the uncertainty bounds in addition to providing dam-
age trajectories (Goebel, Saha, Saxena, & Field, 2008). RVM
(Tipping, 1999) is a Bayesian formalism representing a gener-
alized linear model of identical functional form of the support
vector machine (SVM). Although SVM (Vapnik, 1995) is
a state-of-the-art technique for classification and regression,
RVM is able to generate probabilistic outputs in a Bayesian
framework that make more sense in RUL estimation appli-
cations and futhermore uses a lot kernel functions for com-
parable generalization performance (Goebel, Saha, Saxena,
& Field, 2008). A GP is a collection of random variables,
any finite number of which have a joint Gaussian distribution.
The distribution of a Gaussian process is the joint distribution
of all those (infinitely many) random variables, and as such,
it is a distribution over functions with a continuous domain,
e.g. time or space. In (Goebel, Saha, Saxena, & Field, 2008)
the authors evaluate NN-based approach, RVM and GPR for
their prognostic capabilities on a test stand involving rotat-
ing equipment in an aerospace setting. In the paper, however,
there is no clear winner, since each of the algorithms came up
with its own current state estimates which were not close to
each other. The conclusion states that even though these algo-
rithms can learn the dynamics of the process from sparse and
noisy data fairly well, the RUL estimates depend significantly
on the current state estimation.
Other approaches used are methods from time-series analysis.
The autoregressive moving average (ARMA) model forms a
class of general linear models used in modelling and forecast-
ing of time-series. It is comprised by two parts, namely one
for the autoregression (AR) and the second for the moving
average (MA). It is a powerful forecasting methodology that
is able to capture trends found in a time series and projects its
future values. In a recent paper by Baptista et. al (Baptista
et al., 2018), the authors integrate the ARMA methodology
with data-driven techniques, to predict fault events on a real
industrial case of unscheduled removals of the engine bleed
valve (EBV), based only on life-usage data (maintenance
event data). EBV is used in most designs as a regulator for the
flow that goes to the ECS and the anti-icing systems of the air-
craft. The authors proposed a method in which they feed the
entire past fault event history into the ARMA model and the
output is then used as a feature that integrates with the data-
driven model. The data-driven modeling gives further insight
into the forecasting outcome from ARMA and improve its
accuracy and efficiency. From the data-driven methods they
used, in addition to ARMA (NN, k-nearest neighbors (KNN),
21
INT ERNATIO NAL JO UR NAL O F PROGNOSTICS AND HEALT H MANAGEMENT
random forest (RF), support vector regression (SVR), gener-
alized linear regression (GLM)) the SVM produced the best
overall results. In a similar manner, Su et. al used in (Su,
Zhang, & Zhao, 2014) least squares support vector regres-
sion with sliding ARMA forecasting to model the non-linear
time series. They demonstrate their method on a practical
case study for the US-made F-16 fighter.
However, ARMA models are applied in cases were data show
evidence of a stationary stochastic process. This means that
the time series’ statistical properties are all constant over
time. A stationary series has no trend. That is, its variations
around its mean have a constant amplitude, and it wiggles
in a consistent fashion, i.e., its autocorrelations remain con-
stant over time. Equivalently, short-term random time pat-
terns always look the same in a statistical sense. If the con-
trary stands, its generalization, the Autoregressive Integrated
Moving Average (ARIMA) model can be adopted. The ”Inte-
grated” indicates that the data values have been replaced with
the difference between their values and the previous values, to
transform the time series to a stationary one. In a recent paper
by Ord´
o˜
nez et al. (Ordez, Snchez Lasheras, Roca-Pardias, &
Juez, 2019), the authors combine time series analysis meth-
ods (ARIMA) to forecast the values of the predictor variables
with machine learning techniques to predict RUL of aircraft
engine for more than one period ahead, from those variables.
Another important category of data-driven models used are
graphical models, which denote the conditional independence
structure between random variables (Chen Xiongzi et al.,
2011). In a recent paper, (Banghart, Bian, Strawderman,
& Babski-Reeves, 2017), Banghart et al. utilize Bayesian
networks (BN) to estimate the risk of the landing gear sys-
tem, cockpit warning/caution annunciator panel and the en-
vironmental control system turbine assembly of the Northrop
Grumman EA-6B Prowler military aircraft. BN is a proba-
bilistic graphical model that represents a set of variables and
their conditional dependencies via a directed acyclic graph
(DAG). Nodes represent variables, while arcs represent prob-
abilistic relationships. For example, engine blade damage im-
pacts non mission-capable time, thus an edge/arc is drawn
from the respective nodes. It is a combination of graph the-
ory and probability theory. It is a representation of a joint
probability distribution defined on a finite set of random vari-
ables that can be discrete or continuous. From a knowledge
modeling standpoint, Bayesian networks can be seen as a spe-
cial knowledge representation system. The advantage of BN
lies in the fact that it does not rely on explicit understand-
ing of causal connections within the system(s) under obser-
vation, nor identification of sequences of events leading to
failure. Furthermore, given their probabilistic nature, BNs
prove to be a suitable technique to address the inherent un-
certainty of RUL estimation. In the same view, Ferreiro et
al. in (Ferreiro & Arnaiz, 2010) use BN as a predicting tech-
nique and demonstrate their effectiveness by representing a
physical-model for aircraft brake wear, originally developed
by British Aerospace Systems. They fit it to the available
data (aircraft weight, landing velocity, brake operation dur-
ing landing, flap position and initial brake temperature) about
flight conditions extracted from the operational plan of the
aircraft. Although in this example the causal connections
are based on understanding a physical-system, the general
idea is that BN can be successfully used in prognosis also,
instead of diagnosis. A subclass of BN are the so-called
dynamic bayesian networks (DBN), which relate variables
to each other, over adjacent time steps. They can be con-
sidered simply as BN for the modelling of time-series data
(Ghahramani, 2001). A specification of DBN are the hidden
Markov models (HMM), which have been applied to prog-
nostic problems in aviation.
HMM is a stochastic process model, characterized by a dou-
bly embedded stochastic process with an underlying hidden
stochastic process that can be observed through some prob-
abilistic behaviour. The latter justifies the word hidden. It is
also a powerful tool for RUL estimation. HMM is further-
more a parametric model with some distinct characteristics:
it can not only reflect the randomness of machine behaviour
(i.e., sensor measurements) but also reveal hidden states and
changing processes (Ahmadzadeh & Lundberg, 2014). For a
more thorough understanding we direct the reader to (Paul A.
Gagniuc, 2017). In this view, in (Bechhoefer, Bernhard, He,
& Banerjee, 2006), the authors investigate the use of a Hidden
semi-Markov model (HSMM) to predict the RUL of the shaft
of utility helicopters until failure. The difference between a
HSMM and a HMM is that in the latter the amount of time the
process spends in a state before making a transition is iden-
tical to 1 (for the discrete case) or exponentially distributed
(for the continuous case). Thus, for semi-Markov processes,
an upcoming transition’s distribution is described by a prod-
uct of an arbitrary PDF for the waiting time and a categorical
distribution for the next state. The arbitrary condition for the
PDF, removes the memorylessness property of the process
and as such, the process is Markovian only at the specified
jump instants. Dong et al. in (Dong, He, Banerjee, & Keller,
2006), proposed a HSMM for fault classification application
for UH-60A Blackhawk main transmission planetary carriers
and prognosis of a hydraulic pump health monitoring appli-
cation. They compare HSMM with HMM and conclude that
the former are capable of identifying the faults under both
test cell and on-aircraft conditions while the performance of
the HMM is not comparable with that of the HSMM, while
also the HSMM-based methodology can be used to estimate
the RUL of equipment. However, HMM have some inher-
ent limitations. One is the assumption that successive system
behaviour observations are independent and the other is that
the Markov assumption that the probability in a given state at
time tonly depends on the state at time t−1is clearly un-
tenable in practical applications (Ahmadzadeh & Lundberg,
22
INT ERNATIO NAL JO UR NAL O F PROGNOSTICS AND HEALT H MANAGEMENT
2014).
In the same context, time-series analysis methods, which we
referred to before, have been combined with HMM. Specif-
ically, AR models have been combined with HMM, in what
is called an autoregressive hidden markov model (ARHMM)
(Poritz, 1982), initially proposed for speech recognition.
Here the observations are drawn from an autoregression pro-
cess (linear prediction) (Rabiner, 1990). In this view, Jue-
sas et al. (Juesas, Ramasso, Drujont, & Placet, 2016) devel-
oped a variant of ARHMM, named autoregressive partially-
hidden markov model (ARPHMM) for fault detection and
prognostics of equipment based on sensors’ data. The authors
considered a modification of the learning procedure of the
ARHMM, by integrating prior knowledge on latent variables.
Their method was demonstrated on an instance of the C-
MAPSS dataset. They compared their approach, on the afore-
mentioned dataset, against RULCLIPPER (Remaining Useful
Life estimation based on impreCise heaLth Indicator mod-
eled by Planar Polygons and similarity-basEd Reasoning)
(Ramasso, 2014) (see also Section 8), and SWELM (Summa-
tion Wavelet Extreme Learning Machine) (Javed, 2013). The
results are promising, in the sense that they are comparable
to RULCLIPPER, and, on average, better than SWELM. It is
interesting however, as the authors point out, that using an en-
semble between ARPHMM and SWELM outperforms RUL-
CLIPPER, showing a direction towards developing ensemble
approaches made of complementary and advanced prognos-
tics algorithms.
Finally, a method with which we would like to conclude this
section is the use of Neuro-Fuzzy systems (NF) for prognos-
tics. NF systems are neural network-based fuzzy systems,
with the latter being a nonlinear mapping of an input data
vector with a scalar output. Fuzzy logic is based on Zadeh’s
fuzzy set theory (Kan, Tan, & Mathew, 2015). For a bet-
ter understanding of fuzzy logic and neuro-fuzzy systems, we
refer the reader to (Ross, 2010) and (Abraham, 2005), respec-
tively. In (Chen, Vachtsevanos, & Orchard, 2012), the authors
propose an integrated adaptive neuro-fuzzy inference systems
(ANFIS) and high-order particle filtering, which forecasts the
time evolution of the fault indication and estimates the prob-
ability density function of RUL. The ANFIS is used to model
the fault propagation trend and the high-order PF integrates
the ANFIS, as an m-th-order hidden Markov model, to carry
out long-term predictions and estimate the RUL PDF via a
set of particles with associated weights. They apply their
method on vibration data from the main gearbox of a UH-
60 helicopter subjected to a seeded carrier plate crack fault
and show that its prediction accuracy is higher than that of
both the conventional ANFIS predictor and the particle-filter-
based predictor where the fault growth model is a first-order
model that is trained via the ANFIS.
The advantages of data-driven methods for prognostics are
that they can be deployed quicker and cheaper compared to
other approaches such as model-based. This is true especially
when the monitored system is sufficiently complex, such that
developing an acceptably accurate model is prohibitively ex-
pensive. Another advantage of data-driven methods is their
versatility and cost in terms of development, as well as that
they refer to a wider audience of researchers than model-
based methods. Furthermore, they can provide a system-wide
coverage in comparison to model-based methods, which usu-
ally have a more narrow scope in order to model a particular
system or component. Finally, they are popular due to their
ability to transform high-dimensional noisy data into lower
dimensional information for diagnostic/prognostic decisions
(Dragomir et al., 2009).
In view of this, we are presenting part of (Vatani et al., 2015),
to allow the reader to understand the aforementioned. In
(Vatani et al., 2015) the authors predict the degradation trends
of a gas turbine engine by studying their effects on sensor data
(i.e. temperature) by using an RNN, as one the different ap-
proaches. We will describe this approach here and we refer
the interested reader to the original paper for all the explicit
details and methods.
The authors’ goal is to predict the trend of a single-spool tur-
bofan engine, and for this purpose, identify the major reasons
for engine degradation. They consider the effects of fouling
and erosion, which constitute the most significant long-term
effects on engine performance, when compared to the others,
and study their effects only for take-off, since this is the flight
phase in which the effects and growth of degradation are more
prominent. We will cover here the effects of fouling and its
effects on the compressor as studied by the authors.
For the fouling degradation modeling they define initially a
fouling index (FI), to represent the fouling effects on the en-
gine performance. The FI is modeled as a linear degradation.
Secondly, they introduce a degradation index (DI) to quan-
tify its relationhsip to mass flow capacity and efficiency and
define the rate of change in mass flow capacity of the com-
pressor and its efficieny, as well as the degradation gain in
each cycle of the mass flow capacity and efficiency of the
compressor. It should be pointed out that the objective is not
to train the RNN to identify the dynamics of the engine, but
instead to train the networks to represent, learn and predict
the dynamics of the degradation process on the engine vari-
ables and how it grows in the system as the flight cycles build
up.
For the data generation, they specifically simulate the ground-
roll phase, which includes taxiing until lift-off, and all simula-
tions are performed under standard operating conditions (am-
bient temperature and pressure assumed to be at the standard
sea-level at take-off). The results related to the degradation
modelling and the changes to the engine output parameters
due to different degradation levels are compared with the GSP
23
INT ERNATIO NAL JO UR NAL O F PROGNOSTICS AND HEALT H MANAGEMENT
10 (Gas turbine Simulation Program - National Aerospace
Laboratory NLR) single spool turbojet model. They picked
this phase because the engines operate from their idle posi-
tion to their maximum power and the effects of degradation
are more significant. In the simulations every flight cycle
takes 3000 s, out of which the take-off is 20 s. The com-
pressor pressure and temperature are recorded, as well as the
spool speed and the fuel flow rate. In this work the effects
of the turbine temperature (TT) variations is chosen as a suit-
able indicator, due to the fact that changes in the compressor
health parameters (compressor flow rate and compressor effi-
ciency) affect the gas path measurements. Finally, to manage
the uncertainty of the prognosis, they define upper and lower
bounds. When the upper bound of the prediction variable
meets a specified threshold, one can declare that the engine
should be taken for maintenance.
For the fouling effects they generate data for 1% fouling to
train the RNN. Based also on the assumption that fouling is
a soft degradation and as such, does not change the engine’s
performance in one flight (drastically), they feed the network
with just the 12th second snapshot.
Initially, they consider an RNN with a 4-5-1 architecture i.e.,
4 input nodes, 1 hidden layer with 5 nodes and 1 ouput node.
They then do a 3 step-ahead prediction. The statistical error
results are given as (mean error) µ= 0.177 Kelvin, (standard
deviation of the prediction error) σ= 1.384 Kelvin, and (root
mean squared error) RMSE = 1.343 Kelvin. The results are
shown in figure 3 in the original paper. There, we can see that
most of the TT values lie in the prediction bounds as predicted
by the RNN and only two points fall out of the bands, giving
a 92.6% of points inside. They further consider a 10 steps-
ahead prediction with the 1% fouling using a 6-5-1 architec-
ture, i.e., 6 neurons in the input layer, 5 neurons in the hidden
layer, and 1 neuron in the output layer. The statistical results
for the 10 steps-ahead prediction are (mean error) µ= -0.338
Kelvin, (standard deviation of the prediction error) σ= 1.810
Kelvin, and (root mean squared error) RMSE = 2.786 Kelvin,
which represent a reasonable error level, taking into account
the prediction horizon. The results are shown in figure 4 in the
original paper. In the figure one can observe that several ac-
tual (measured) data points have exceeded beyond the bounds
and only 65% lie inside it. They conclude that the achieved
error level for the second RNN meets the expected specifica-
tions as the maximum error is 4 degrees Kelvin which is less
than 1% of the range of 1400K, implying that the prediction
is 99% accurate. They also conclude that the first RNN is
the most suitable network for predicting a 1% fouling effect
on the TT. Finally, with the prediction results one can obtain
the system RUL when the gas path measurements reach pre-
specified thresholds.
To conclude this section, we must also point out some disad-
vantages of data-driven methods. One important is that they
are data-hungry, that is, they require a substantial amount of
data for training and thus, they are highly-dependent on the
quantity and quality of the system’s operational data. There
is also commonly a lack of a procedure to obtain the training
data and there is further a lack of run-to-failure data, for the
methods to learn from. Due to this, applications in the liter-
ature just use experimental data for model training and thus
these approaches may have wider confidence intervals than
others. (Dragomir et al., 2009).
To counter these limitations, including the limitations of
model-based techniques, fusion approaches have been de-
signed, to combine their strengths. In the next section, we
discuss these.
7.6. Applications of Prognostics Hybrid Methods in the
Aerospace Industry
Fusion or hybrid-based prognostic methodologies combine
the strengths of the model - based and data - driven ap-
proaches, in order to estimate the RUL under both operating
and non-operating life cycle conditions (Pecht & Jaai,
2010). By taking advantage of the two respective methods’
strengths, hybrid models can achieve robust health prediction
results that can lead to a more reliable RUL approximation
as compared to only model-based methods. Also, due to the
use of a mathematical model, the amount of data required
for training purposes are relatively lower than that needed
in pure data - driven methods (Daroogheh, Baniamerian,
Meskin, & Khorasani, 2015). A hybrid model thus combines
both data-driven methodologies with the knowledge of the
system under study.
In a recent paper by Daroogheh et al., (Daroogheh et
al., 2015), the authors proposed a hybrid structure for
the health monitoring of non-linear systems. Specifically,
they fuse particle filtering to estimate the states and health
parameters of the system and extend the PF to the future time
horizon by utilizing a neural network (NN) as a non-linear
forecasting scheme. The authors apply the proposed method
to predict the health condition of a gas turbine engine.
Also recently in (Xu, Wang, & Xu, 2014) the authors pro-
posed a PHM-oriented hybrid prognostics framework for air-
craft engines based on the measured sensor data, with a fo-
cus on RUL prediction. They are advocates of the idea that
for such complicated problems it is difficult to model the
complex degradation process, and no single prognostic ap-
proach can effectively solve this. Their fusion prognostic
method utilizes the aircraft engine sensor data and combines
the strengths of both data-driven and experience-based prog-
nostic approaches and eliminates their limitations. In the
same view (Xu & Xu, 2011) design a fusion prognostic model
for avionics systems by combining data-driven, model-based
and experience-based PHM methods. Their fusion model is
24
INT ERNATIO NAL JO UR NAL O F PROGNOSTICS AND HEALT H MANAGEMENT
built upon the optimal linear combination forecast model us-
ing absolute value and prediction error as the index of prog-
nostic precision. They conclude that their fusion prognostics
can provide RUL estimation more accurate and more robust
than either of the algorithms alone (ARMA, SVM, FNN).
In (Byington, Watson, & Edwards, 2004), the authors de-
signed a prognostics framework for aircraft actuator compo-
nents. Their methodology includes automated signal process-
ing and neural network tracking techniques, along with auto-
mated reasoning, classification, knowledge fusion, and prob-
abilistic failure mode progression algorithms. They imple-
ment a fuzzy logic process to quantify the level of damage
(damage index) of the system using a predefined set of rules
tailored from knowledge of the system and engineering judg-
ment (experience-based method). A physics-based method
(Kalman filtering) was then applied to predict the progression
of the damage. They demonstrated their method to F/A-I8
stabilator electro-hydraulic servo valves (EHSVs).
In (Heimes, 2008), Heimes, applied an RNN trained by back-
propagation method, using extended Kalman filter (EKF)
training method to update the weights of the network. The
EKF is particularly attractive for RNN training, since it min-
imizes the number of training iterations and does not require
all the training data to be utilized for each training iteration.
The RUL for the simulated aircraft engines was calculated by
simply averaging the outputs of the three best models. This
solution won 2nd place in the IEEE 2008 PHM Conference
challenge problem (see (Ramasso & Saxena, 2014)), in which
the contestors where asked to estimate the RUL (remaining
operational cycles) of multiple multivariate time-series of en-
gine data.
In (Howard, Mesick, Reuter, & Roemer, 2001), the authors
present an architecture that combines the model-based and
data-driven approaches for fault detection, diagnosis, and
prognostics for aircraft. It includes a prognostic reasoner that
takes as input the outputs from a variety of specialized prog-
nostic algorithms for different systems within the aircraft, and
then prioritizes the most probable failure modes. The system
uses an integrated model to predict how a probable failure
will affect systems throughout the aircraft. It also compares
the outputs of the anomaly detector, the diagnostic system,
and the prognostic system to see whether the three types of
algorithms endorse each other. The authors also include a
brief description of some fault detection and diagnostic algo-
rithms that they have used with the architecture, but there is
no description of prognostic algorithms.
In (Lasheras, Nieto, de Cos Juez, Bayn, & Surez, 2015) the
authors use a hybrid model, with the meaning that it is usu-
ally employed in the field of machine learning and pattern
recognition, for predicting the RUL of aircraft engines. They
combine the multivariate adaptive regression splines (MARS)
technique with the principal component analysis (PCA), den-
drograms and classification and regression trees (CARTs).
The proposed model does not require information about the
previous status of the variables of the engine and it only re-
quires information regarding the current situation of the stud-
ied variables and is trained by elements extracted from the
engine’s sensors representing different levels of health for air-
craft engines. They conclude that their model performs better
than the classical predictive models (linear multivariate re-
gression model and neural network model with back propa-
gation) applied in recent years for the modeling of RUL.
We decided to present the work of (Daroogheh et al., 2015) to
give the reader an example of how a fusion method can work.
In this paper the authors apply a hybrid method to estimate
the RUL of a compressor of a single spool jet engine, which is
being degraded due to fouling phenomena. The single spool
jet engine model they used was developed in (Naderi, Me-
skin, & Khorasani, 2011), (Naderi, Meskin, & Khorasani,
2012). Specifically they combine state/parameter estimation
using particle filters (PF) (model-based method) with neural
networks for observation prediction (data-driven method).
For the model-based part, they implement two parallel fil-
ters for the state and parameter estimation tasks. At each
time step for the state (parameter) estimator filter, the cor-
responding parameters (states) are treated as being a known
input to the filter from the parameter (state) estimator. Re-
garding the data-driven part, for the NN, they used a multi-
layer perceptron (MLP) and RNN and wavelet neural network
(WNN). With the use of the NN, they forecasted observa-
tions/measurements (data-driven prediction) allowing for the
PF algorithm to be extended to future steps by utilizing the
same weight update rule used in the estimation step. For this
task, they use the mean squared error (MSE) when new ob-
servations are available, in order to validate and if, needed,
retrain the network if the MSE exceeded a predefined thresh-
old.
For their application they use as system state parameters, the
compressor and turbine efficiency and the compressor and
turbine mass flow capacities. They apply a linear degrada-
tion model on the compressor health parameters during 1000
cycles of operation causing a drop of 3% to compressor effi-
ciency and 1.5% to its mass flow capacity and they define as
the critical values 0.7832 for the compressor efficiency and
20.0127 kg/s for the compressor mass-flow capacity. The au-
thors use as a ground truth failure cycle (based on the afore-
mentioned critical values) to be at 230 and predict 40 steps
ahead from cycle 200. The latter is because they assume that
the required data to train the networks are 200. The results
can be seen in the original paper in figure 3 and figure 4.
The presented results indicate that the PF method without in-
tegration (thus a standalone implementation) with an obser-
vation forecasting module (such as the implemented NN), is
not able to track the changes in the compressor efficiency cor-
25
INT ERNATIO NAL JO UR NAL O F PROGNOSTICS AND HEALT H MANAGEMENT
rectly and has overestimated this health parameter. It is ap-
parent that it cannot be located inside the +/-99% confidence
interval around the real value. Although for mass flow capac-
ity (figure 3 bottom of the original paper) of the compressor,
the PF method has estimated the value of this health parame-
ter inside the confidence bound for most of the times, the di-
rection of the changes in predicted mass flow capacity in the
fouling scenario is not correct. However, it is evident that all
of the three neural networks are able to predict the compressor
health parameters and lie within the +/-99% confidence inter-
val around the actual values. In figure 4 of the original paper
the distribution of the health parameters is presented, along
with the cycle at which the maximum failure probability oc-
curs. They propose that the time at which the first parameter
associated with the specific degradation damage reaches its
critical value, is the failure cycle. In this example clearly the
compressor mass flow capacity is the parameter that first ex-
ceeds its critical value. The predicted failure value is cycle
215 (RUL=15 cycles) for the RNN, 220 (RUL=10 cycles) for
the WNN and 222 (RUL=8 cycles) for the MLP.
We hope that this paper made it clear to the reader how differ-
ent methods can be combined and that there is a benefit when
using fusion methods in prognostics, especially for non-linear
dynamical systems, such as gas turbine engines.
With the hybrid methods, we conclude the review on PHM
methodology in the aerospace industry. In the next section
we will conclude with this chapter of this literature review.
7.7. Discussions and Conclusions
With the number of passengers being carried through air
transport (both domestic and international), estimated at
3.979 billion in 2017 (International Civil Aviation Organi-
zation & staff estimates., n.d.), safety and economy for air-
craft manufacturers and operators are of greatest significance.
Prognostics and Health Management (PHM) allows for this,
by performing diagnostics and prognostics on the engines,
airframe and other components of an aircraft. The identifi-
cation and isolation of a fault as well as the predictive diag-
nostics give the OEMs and airlines the ability to identify and
recover from failures and provide them with a heads-up on
what to expect. The latter is the essence of PHM and what
has directed the transition from traditional health monitoring
to health management. Since it foresees a fault or failure, it
allows for maximal availability and reliability, shifting away
from the time-scheduled maintenance to a condition-based
maintenance, thus increasing safety and reducing unneces-
sary downtime. These early warnings grant the airline also
the ability to make a timely maintenance strategy schedule
and follow all the necessary procedures for the logistics, such
as ordering parts and making agreements with external ven-
dors.
Being around for only a few decades, especially in the
aerospace industry, prognostics has still plenty of research
potential in this field. It is advisable for future investigations
that a design of a standard procedure for obtaining data for
PHM is developed. Whether a researcher or an airline is using
a model-based a data-driven or a fusion approach, measure-
ment data are needed and are crucial for most methods. It is
the view of the authors, that data science should play an indis-
pensable role in such industries. In addition, there is a lack of
run-to-failure data for the data-driven algorithms to train on
and validate the results. Future research should thus empha-
size also on generating data for such purposes, taking into ac-
count that, especially in this industry theoretical predictions
and methods developed must be verified and validated first
before deployed in practical applications.
8. GENERAL DISCUSSIONS AND CON CL US IO NS
Prognostics and health management (PHM) is a fairly new
discipline that provides actionable information, enabling in-
telligent decision making for improved performance, safety,
reliability, and maintainability of engineered systems, by us-
ing real time and historical state information of subsystems
and components. Aside from that, its importance lies in
that these early warnings grant the user the ability to make
a timely maintenance strategy schedule and follow all the
necessary procedures for the logistics/supply-chain. It differs
from diagnostics in that it can be interpreted as the process of
health assessment and prediction, which includes detecting
incipient failure and predicting remaining useful life (RUL).
It, however, naturally embodies diagnostics which identifies
and determines the relationship between the failure mecha-
nism and the failure mode.
In this paper, we have reviewed the recent researches and de-
velopments that have been done to apply PHM in the auto-
mobile and the aerospace industry. We initially approach the
matter generally by presenting the discipline and its signifi-
cance, before moving on to the applications and methods as
they are specifically used in these industries, focusing on the
remaining useful life (RUL) estimation.
The main approaches adopted by PHM experts fit into four
broad categories. Namely, reliability-based, model-based,
data-driven and hybrid or fusion methods. In this review
we covered the latter three. Model-based methods assume a
mathematical model of the monitored system and use residu-
als as features. Residuals are the outcomes of the compar-
isons between the measurements of a real system and the
outputs of a mathematical model of that system. Due to the
physical understanding of the system, model-based methods
are very accurate. However, this close relation to the math-
ematical model can make them impossible to design, due to
the high complexity. Data-driven methods on the other hand,
rely on data routinely recorded from the monitored systems
and as such they are operational data. Thus, in such cases
26
INT ERNATIO NAL JO UR NAL O F PROGNOSTICS AND HEALT H MANAGEMENT
in which data are available the data-driven methods are pre-
ferred because of their ease of use These methods adopt tech-
niques from AI, machine learning and pattern recognition.
Their main advantage compared to model-based methods, is
the fact that they can be deployed faster and cheaper, since
they do not assume anything about the underlying relation-
ships of the system under study and therefore they appeal to a
wider audience and techniques. Everything comes with a cost
though. These methods might be cheaper to deploy, but rely
largely on data, which is most often of questionable quality,
is rare or not existent or cannot be disclosed. Also, their accu-
racy might not be as high as that of a model-based approach.
Hybrid approaches combine model-based and data-driven ap-
proaches strengths while at the same time minimizing their
disadvantages. Hybrid approaches are newly developing and
promising. However, to the best of our knowledge, except for
battery, hybrid approaches have not been widely used in PHM
for the automobile- and aerospace industries. We particularly
addressed the importance of performance metrics and uncer-
tainty evaluation in PHM. Uncertainties arise from various
sources such as: modeling uncertainties, measurement uncer-
tainties, operating environment uncertainties and others alike.
Such information is crucial for any prognostic approach and
must be validated before being incorporated in PHM applica-
tions.
Despite the fact that PHM has been around for just a few
decades, plenty of work has been done in estimating the RUL.
However, their practical success seems to be still limited in di-
agnosis, because there are a number of challenges and prac-
tical issues to be addressed. For example, even though re-
searchers developed data-driven methods to counter the in-
creasing complexity of systems and components, there is still
no standard procedure to obtain data, in terms of a protocol
or system. Data are either not integrated centrally, but scat-
tered around different systems, or cannot be disclosed due to
security and privacy issues and competition. To add to that,
there is also the possibility that data do not even exist due
to cost, as for example run-to-failure data of a turbofan en-
gine. Future research should thus emphasize also in generat-
ing data for such purposes or when it is not available, taking
into account that in field applications theoretical predictions
and methods developed must be verified and validated first
before practical applications become possible. Finally, for
fielded applications another crucial challenge is the real-time
(online) RUL estimation. The issue lies in the fact that the
developed methods need intensive computational resources,
which is in direct contradiction with hardware conditions of
on-board computers.
What is more, setting aside the fact that plenty of fundamen-
tal work has been done in the context of PHM, the issue still
remains that simplifying assumptions have to be made, since
the underlying system is too complex. For example, in model-
based, simplifying assumptions are made about the system, in
order to be able to mathematically model it. In data-driven, on
the other hand, assumptions are made regarding i.e., the shape
of the health degradation curve, and in addition results are not
always interpretable. While these approaches have given us
remarkable results, and still have a lot to offer, we feel the
urge to suggest future work towards an area that, we believe,
has been overlooked in prognostics, and can overcome some
of the shortcomings mentioned above, for the data-driven and
model-based approaches. That method is evolutionary com-
putation (EA) and particularly symbolic regression (SR). SR
is a subclass of genetic programming (GP), which in turn is
a subclass of genetic algorithms (GA), that belong to evo-
lutionary algorithms (EA). SR is a regression technique that
searches the space of algebraic expressions, to find the model
that best fits the given data, without making any assumptions
about the structure of the model (i.e., linear or non-linear). It,
instead, evolves mathematical formulas, in the evolutionary
computation sense. Such an approach can help mitigate some
of the issues mentioned above. For example it can be very
effective in uncovering the physics that govern a physical-
system, such as a turbofan engine, and as a result increases
system understanding. In such a case sensor measurements
provide us with I/O data. A SR could unveil the underly-
ing mathematical relations, which could then be used as a
model-based approach in the standard way, as discussed in
this review. Furthermore, by uncovering the generating pro-
cess of the data, it can provide insight on the quality of the
data recorded. What is more, it can alleviate the black-box
approach of usual data-driven methods, since it provides a
mathematical expression as a result, and thus allowing the
model to be better interpreted.
In addition, most of developed PHM approaches follow the
modern control theory (MCT); meaning the established mod-
els (mathematical, physical or machine learning models) are
built on the train datasets then they are validated on the test
datasets. Normally, system identification algorithms can be
employed to train a model from data, which is then used to
design the controller (models and hyper-parameters). De-
signing a controller by first identifying an open-loop model
of the process, and then synthesizing a controller based on
the resulting model, would require focusing the attention on
reproducing the open-loop behavior of the process first, then
selecting a model structure, tune the model parameters, com-
pare the open-loop response of the model on validation data,
etc (Selvi, Piga, & Bemporad, 2018). Therefore, the cho-
sen established models should reflect the desired closed-loop
performance in a PHM system and be reproducible by the
underlying unknown process when in a closed-loop with the
synthesized controller. However, building a model in PHM
is neither costly- nor time-consuming, it remains difficult to
decide a-priori the level of accuracy/complexity the model
should have to meet the desired closed-loop performance.
Difficulty in modeling, poor robustness, lack of safety, as well
27
INT ERNATIO NAL JO UR NAL O F PROGNOSTICS AND HEALT H MANAGEMENT
as a huge gap between theoretical analysis and practical per-
formance due to un-modeled dynamics and parametric uncer-
tainty, are common phenomena when applying control theory
methods in practice (Hou & Jin, 2014).
Therefore, using data-driven model-free (DDMF) methods
would be a new trend in PHM. In these methods, the mea-
surement input/output (I/O) data is directly used for control
design and closed-loop system analysis, without any model
dynamics involved, thus, DDMF is also called as data-driven
control method. (Hou & Jin, 2014). In DDMF, adaptive
control for an unknown nonlinear system with time-varying
parameters and time-varying structure is uniformly realized,
and the existing difficulties in modern control methods, such
as the dependence of controller design on system model, un-
modeled dynamics, traditional robustness issues, and other
related theoretical problems, are avoided within the data-
driven control framework. Several examples of the DDMF
frameworks are Virtual Reference Feedback Tuning (VRFT)
(Campi, Lecchini, & Savaresi, 2002), Iterative Feedback Tun-
ing (IFT) (Hjalmarsson, 2002), Simultaneous Perturbation
Stochastic Approximation (SPSA) (Spall & Cristion, 1998),
Model-Free Iterative Learning Control (MFILC) (Radac, Pre-
cup, & Petriu, 2015), Model-Free Adaptive Control (Hou &
Jin, 2011; Fliess & Join, 2013).
Both the theoretical results (Radac, Precup, & Roman, 2018;
Selvi et al., 2018; Guo, Hou, Liu, & Jin, 2019) incessantly
developed and improved in the past two decades, and their
promising practical applications in chemical industry (dos
Santos Coelho, Pessˆ
oa, Sumar, & Coelho, 2010), healthcare
(Meneghetti, Terzi, Del Favero, Susto, & Cobelli, 2018), ma-
chinery (Tan, Lee, Huang, & Leu, 2001; Radac & Precup,
2017; Duan, Hou, Yu, Jin, & Lu, 2019; Stainier, Leygue, &
Ortiz, 2019), and so on, would suggest DDMF to be novel
methods for developing systematic and rigorous PHM frame-
works. As an outstanding example, in 2014, Ramasso es-
tablished a RULCLIPPER algorithm for estimating the RUL
of physical systems, by utilising health index/health indicator
(HI), which reflect the system’s health state. In this DDMF
framework, the HI are assumed imprecise (IHI) (Ramasso,
2014). RULCLIPPER is comprised of ideas inspired from
both computer vision and computational geometry, with IHI
data interpreted as vertices of a simple (non self-intersecting)
planar polygon. The algorithm relies further on the adaptation
of case-based reasoning (see (T. Wang, Jianbo Yu, Siegel, &
Lee, 2008)) to manage the imprecise training and testing in-
stances. The method was validated with C-MAPSS datasets
and illustrated that computational geometry seems promising
for PHM in presence of noisy HIs. While some similarity-
based matching algorithms (case-based reasoning) may suf-
fer from computational complexity, in particular to sort train-
ing instances, the proposed algorithm is shown to be efficient
with few parameter tuning on all datasets. Additionally, the
authors claim that RULCLIPPER is a general algorithm, in-
variant to the peculiarities of the datasets that it can be used
on.
To conclude, future directions for research in PHM should
include the use of data science tools for generating, and ef-
fectively disseminating data. This includes also the integra-
tion of data sources with smart technologies for the end-users.
What is more, there is a need for research in uncertainty rep-
resentation for online evaluation of prognostics to deal with
uncertainties associated with future operating conditions, in
order for prognostics to be deployed in the real world.
We surely hope that this review will help researchers in the
field to get an overview on recent developments of PHM in
the aerospace and automobile industry, as well as a more gen-
eral picture of PHM and its importance. We believe that PHM
related research will continue to grow in the following years,
as there is still much to learn on the potential of data, as well
as to mitigate all the current shortcomings.
ACKNOWLEDGEMENTS
This work is part of the research programme Smart Indus-
try SI2016 with project name CIMPLO and project number
15465, which is partly financed by the Netherlands Organisa-
tion for Scientific Research (NWO).
REFERENCES
Abraham, A. (2005, July). Adaptation of Fuzzy Infer-
ence System Using Neural Learning. In J. Kacprzyk,
N. Nedjah, & L. d. Macedo Mourelle (Eds.), Fuzzy
Systems Engineering (Vol. 181, pp. 53–83). Berlin,
Heidelberg: Springer Berlin Heidelberg. doi:
10.1007/11339366 3
Accident: Southwest B737 near Philadelphia on Apr 17th
2018, uncontained engine failure takes out passenger
window. (n.d.).
Ahmadzadeh, F., & Lundberg, J. (2014, December). Re-
maining useful life estimation: review. International
Journal of System Assurance Engineering and Man-
agement,5(4), 461–474. doi: 10.1007/s13198-013-
0195-0
Alia, J. B., Chebel-Morello, B., Saidi, L., Malinowski, S., &
Fnaiech, F. (2015). Accurate bearing remaining useful
life prediction based on weibull distribution and artifi-
cial neural network. Mech Syst Signal Process, 150-
172.
Arulampalam, M. S., Maskell, S., Gordon, N., & Clapp,
T. (2002). A Tutorial on Particle Filters for Online
Nonlinear/Non-Gaussian Bayesian Tracking. IEEE
Transactions on Signal Processing,50(2), 15.
Ashok, B., Denis Ashok, S., & Ramesh Kumar, C. (2016).
28
INT ERNATIO NAL JO UR NAL O F PROGNOSTICS AND HEALT H MANAGEMENT
A review on control system architecture of a SI engine
management system. Annual Reviews in Control,41,
94–118. doi: 10.1016/j.arcontrol.2016.04.005
Ashok, B., Denis Ashok, S., & Ramesh Kumar, C.
(2017). Trends and future perspectives of elec-
tronic throttle control system in a spark ignition en-
gine. Annual Reviews in Control,44, 97–115. doi:
10.1016/j.arcontrol.2017.05.002
Atamuradov, V., Medjaher, K., Dersin, P., Lamoureux, B.,
& Zerhouni, N. (2017). Prognostics and health man-
agement for maintenance practitioners - Review, im-
plementation and tools evaluation. , 32.
Banghart, M., Bian, L., Strawderman, L., & Babski-
Reeves, K. (2017, July). Risk assessment on
the EA-6b aircraft utilizing Bayesian networks.
Quality Engineering,29(3), 499–511. doi:
10.1080/08982112.2017.1319957
Baptista, M., Sankararaman, S., de Medeiros, I. P., Nasci-
mento, C., Prendinger, H., & Henriques, E. M. (2018,
January). Forecasting fault events for predictive main-
tenance using data-driven techniques and ARMA mod-
eling. Computers & Industrial Engineering,115, 41–
53. doi: 10.1016/j.cie.2017.10.033
Baraldi, P., Compare, M., Sauco, S., & Zio, E.
(2013). Ensemble neural network-based parti-
cle filtering for prognostics. Mechanical Systems
and Signal Processing,41(1), 288 - 300. doi:
https://doi.org/10.1016/j.ymssp.2013.07.010
Beatrice, C., Guido, C., Napolitano, P., Iorio, S. D., &
Giacomo, N. D. (2011, May). Assessment of
biodiesel blending detection capability of the on-
board diagnostic of the last generation automotive
diesel engines. Fuel,90(5), 2039–2044. doi:
10.1016/j.fuel.2011.01.013
Bechhoefer, E., Bernhard, A., He, D., & Banerjee, P. (2006).
Use of Hidden Semi-Markov Models in the Prognostics
of Shaft Failure. , 7.
Berecibar, M., Gandiaga, I., Villarreal, I., Omar, N.,
Van Mierlo, J., & Van den Bossche, P. (2016, April).
Critical review of state of health estimation methods
of Li-ion batteries for real applications. Renewable
and Sustainable Energy Reviews,56, 572–587. doi:
10.1016/j.rser.2015.11.042
Bolander, N., Qiu, H., Eklund, N., Hindle, E., & Rosenfeld,
T. (2009). Physics-based Remaining Useful Life Pre-
diction for Aircraft Engine Bearing Prognosis. , 12.
Brier G.W., & Allen R.A. (1951). Verification of Weather
Forecasts. Boston, MA: In: Malone T.F. (eds) Com-
pendium of Meteorology. American Meteorological
Society.
Byington, C., Watson, M., & Edwards, D. (2004). Data-
driven neural network methodology to remaining life
predictions for aircraft actuator components. In 2004
IEEE Aerospace Conference Proceedings (IEEE Cat.
No.04th8720) (Vol. 6, pp. 3581–3589). Big Sky, MT,
USA: IEEE. doi: 10.1109/AERO.2004.1368175
Campi, M., Lecchini, A., & Savaresi, S. (2002). Virtual refer-
ence feedback tuning: a direct method for the design of
feedback controllers. Automatica,38(8), 1337 - 1346.
doi: https://doi.org/10.1016/S0005-1098(02)00032-8
Celaya, J. R., Saha, B., & Wysocki, P. F. (2009). Prognostics
for Electronics Components of Avionics. , 7.
Chen, C., Vachtsevanos, G., & Orchard, M. E. (2012,
April). Machine remaining useful life predic-
tion: An integrated adaptive neuro-fuzzy and high-
order particle filtering approach. Mechanical Sys-
tems and Signal Processing,28, 597–607. doi:
10.1016/j.ymssp.2011.10.009
Chen Xiongzi, Yu Jinsong, Tang Diyin, & Wang Yingxun.
(2011, August). Remaining useful life prognos-
tic estimation for aircraft subsystems or compo-
nents: A review. In IEEE 2011 10th Interna-
tional Conference on Electronic Measurement & In-
struments (pp. 94–98). Chengdu, China: IEEE. doi:
10.1109/ICEMI.2011.6037773
Cheng, S., Azarian, M. H., & Pecht, M. G. (2010, June). Sen-
sor Systems for Prognostics and Health Management.
Sensors,10(6), 5774–5797. doi: 10.3390/s100605774
Coble, J., Ramuhalli, P., Bond, L. J., Hines, J. W., & Ipad-
hyaya, B. (2015). A Review of Prognostics and Health
Management Applications in Nuclear Power Plants. ,
24.
Company, T. (2006). Benefits of IVHM: An Analyt-
ical Approach. In 2006 IEEE Aerospace Confer-
ence (pp. 1–9). Big Sky, MT, USA: IEEE. doi:
10.1109/AERO.2006.1656072
Dang, X., Yan, L., Xu, K., Wu, X., Jiang, H., &
Sun, H. (2016, January). Open-Circuit Voltage-
Based State of Charge Estimation of Lithium-ion
Battery Using Dual Neural Network Fusion Battery
Model. Electrochimica Acta,188, 356–366. doi:
10.1016/j.electacta.2015.12.001
Daroogheh, N., Baniamerian, A., Meskin, N., & Khorasani,
K. (2015, June). A hybrid prognosis and health mon-
itoring strategy by integrating particle filters and neu-
ral networks for gas turbine engines. In 2015 IEEE
Conference on Prognostics and Health Management
(PHM) (pp. 1–8). Austin, TX, USA: IEEE. doi:
10.1109/ICPHM.2015.7245020
Daroogheh, N., Meskin, N., & Khorasani, K. (2013, June).
Particle filtering for state and parameter estimation in
gas turbine engine fault diagnostics. In 2013 Ameri-
can Control Conference (pp. 4343–4349). Washington,
DC: IEEE. doi: 10.1109/ACC.2013.6580508
Daroogheh, N., Meskin, N., & Khorasani, K. (2014, June).
A novel particle filter parameter prediction scheme for
failure prognosis. In 2014 American Control Confer-
ence (pp. 1735–1742). Portland, OR, USA: IEEE. doi:
29
INT ERNATIO NAL JO UR NAL O F PROGNOSTICS AND HEALT H MANAGEMENT
10.1109/ACC.2014.6859021
Dishant, E. S., Er.Parminder Singh. (2017, April). Sus-
pension systems: A review. International Research
Journal of Engineering and Technology (IRJET). doi:
10.1007/s00521-017-2986-8
Dong, M., & He, D. (2007). A segmental hidden semi-
markov model (hsmm)-based diagnostics and prognos-
tics framework and methodology. Mechanical Sys-
tems and Signal Processing,21(5), 2248 - 2266. doi:
https://doi.org/10.1016/j.ymssp.2006.10.001
Dong, M., He, D., Banerjee, P., & Keller, J. (2006, Octo-
ber). Equipment health diagnosis and prognosis using
hidden semi-Markov models. The International Jour-
nal of Advanced Manufacturing Technology,30(7-8),
738–749. doi: 10.1007/s00170-005-0111-0
dos Santos Coelho, L., Pess ˆ
oa, M. W., Sumar, R. R., &
Coelho, A. A. R. (2010). Model-free adaptive con-
trol design using evolutionary-neural compensator. Ex-
pert Systems with Applications,37(1), 499 - 508. doi:
https://doi.org/10.1016/j.eswa.2009.05.042
Downey, A., Lui, Y.-H., Hu, C., Laflamme, S., & Hu,
S. (2019, February). Physics-based prognostics of
lithium-ion battery using non-linear least squares with
dynamic bounds. Reliability Engineering & System
Safety,182, 1–12. doi: 10.1016/j.ress.2018.09.018
Dragomir, O. E., Gouriveau, R., Dragomir, F., Minca, E., &
Zerhouni, N. (2009, August). Review of prognos-
tic problem in condition-based maintenance. In 2009
European Control Conference (ECC) (pp. 1587–1592).
Budapest: IEEE. doi: 10.23919/ECC.2009.7074633
Drappier, C. J. (2008). A380: Challenges for the Future. ,
34.
Duan, L., Hou, Z., Yu, X., Jin, S., & Lu, K. (2019).
Data-Driven Model-Free Adaptive Attitude Control
Approach for Launch Vehicle With Virtual Reference
Feedback Parameters Tuning Method. IEEE Access,7,
54106–54116. doi: 10.1109/ACCESS.2019.2912902
Ekwaro-Osire, Stephen, Alemayehu, Fisseha M, & Carlos
Gonalves, Aparecido. (2017). Probabilistic Prog-
nostics and Health Management of Energy Systems.
Springer International Publishing.
Elattar, H. M., Elminir, H. K., & Riad, A. M. (2016, June).
Prognostics: a literature review. Complex & Intelli-
gent Systems,2(2), 125–154. doi: 10.1007/s40747-
016-0019-3
Ferreiro, S., & Arnaiz, A. (2010). Prognostics applied to air-
craft line maintenance: Brake wear prediction based on
Bayesian Networks. IFAC Proceedings Volumes,43(3),
146–151. doi: 10.3182/20100701-2-PT-4012.00026
Fleming, W. (2001, December). Overview of automotive
sensors. IEEE Sensors Journal,1(4), 296–308. doi:
10.1109/7361.983469
Fliess, M., & Join, C. (2013). Model-free control. Inter-
national Journal of Control,86(12), 2228-2252. doi:
10.1080/00207179.2013.810345
Ghahramani, Z. (2001). AN INTRODUCTION TO HID-
DEN MARKOV MODELS AND BAYESIAN NET-
WORKS. Hidden Markov Models, 34.
Goebel, K., Saha, B., Saxena, A., Celaya, J., & Christo-
phersen, J. (2008, August). Prognostics in Bat-
tery Health Management. IEEE Instrumentation
& Measurement Magazine,11(4), 33–40. doi:
10.1109/MIM.2008.4579269
Goebel, K., Saha, B., Saxena, A., & Field, M. (2008).
A Comparison of Three Data-Driven Techniques For
Prognostics. , 14.
Gorjian, N., Ma, L., Mittinty, M., Yarlagadda, P., & Sun, Y.
(2010). A review on degradation models in reliability
analysis. In D. Kiritsis, C. Emmanouilidis, A. Koro-
nios, & J. Mathew (Eds.), Engineering Asset Lifecycle
Management (pp. 369–384). London: Springer Lon-
don. doi: 10.1007/978-0-85729-320-642
Guo, Y., Hou, Z., Liu, S., & Jin, S. (2019). Data-Driven
Model-Free Adaptive Predictive Control for a Class of
MIMO Nonlinear Discrete-Time Systems With Stabil-
ity Analysis. IEEE Access,7, 102852–102866. doi:
10.1109/ACCESS.2019.2931198
He, W., Williard, N., Chen, C., & Pecht, M. (2014a). State
of charge estimation for li-ion batteries using neural
network modeling and unscented kalman filter-based
error cancellation. International Journal of Electri-
cal Power and Energy Systems,62, 783 - 791. doi:
https://doi.org/10.1016/j.ijepes.2014.04.059
He, W., Williard, N., Chen, C., & Pecht, M. (2014b). State
of charge estimation for li-ion batteries using neural
network modeling and unscented kalman filter-based
error cancellation. International Journal of Electri-
cal Power & Energy Systems,62, 783 - 791. doi:
https://doi.org/10.1016/j.ijepes.2014.04.059
Heimes, F. O. (2008, October). Recurrent neural networks
for remaining useful life estimation. In 2008 Inter-
national Conference on Prognostics and Health Man-
agement (pp. 1–6). Denver, CO, USA: IEEE. doi:
10.1109/PHM.2008.4711422
Hjalmarsson, H. (2002). Iterative feedback tuningan
overview. International Journal of Adaptive Con-
trol and Signal Processing,16(5), 373-395. doi:
10.1002/acs.714
Holzer, W. (2011). A380 Advanced Cabin Line Maintenance.
, 41.
Hou, Z., & Jin, S. (2011, Nov). A novel data-driven control
approach for a class of discrete-time nonlinear systems.
IEEE Transactions on Control Systems Technology,
19(6), 1549-1558. doi: 10.1109/TCST.2010.2093136
Hou, Z., & Jin, S. (2014). Model Free Adaptive Control. ,
59.
Howard, L., Mesick, J., Reuter, R., & Roemer, M. (2001).
An Evolvable Tri-Reasoner IVHMl System. , 15.
30
INT ERNATIO NAL JO UR NAL O F PROGNOSTICS AND HEALT H MANAGEMENT
IATA. (2018, June). Fact Sheet Fuel. IATA.
Incident: France A388 over Greenland on Sep 30th 2017,
uncontained engine failure, fan and engine inlet sepa-
rated. (n.d.).
International Civil Aviation Organization, C. A. S. o. t. W.,
& staff estimates., I. (n.d.). Air passengers carried
include both domestic and international aircraft pas-
sengers of air carriers registered in the country. (Tech.
Rep.).
Itier, J.-B. (2007). ARTIST2 IMA & ADCN. , 45.
Jafari, M., Khan, K., & Gauchia, L. (2018, December). Deter-
ministic models of Li-ion battery aging: It is a matter
of scale. Journal of Energy Storage,20, 67–77. doi:
10.1016/j.est.2018.09.002
Jaoude, A. A. (2015). Analytic and linear prognostic model
for a vehicle suspension system subject to fatigue. Sys-
tems Science & Control Engineering,3(1), 81-98. doi:
10.1080/21642583.2014.987359
Jardine, A. K., Lin, D., & Banjevic, D. (2006, October). A
review on machinery diagnostics and prognostics im-
plementing condition-based maintenance. Mechanical
Systems and Signal Processing,20(7), 1483–1510. doi:
10.1016/j.ymssp.2005.09.012
Javed, G. R. Z. N., K. (2013, November). Novel fail-
ure prognostics approach with dynamic thresholds for
machine degradation. In IECON 2013 - 39th Annual
Conference of the IEEE Industrial Electronics Soci-
ety (pp. 4404–4409). Vienna, Austria: IEEE. doi:
10.1109/IECON.2013.6699844
Jeong, J., Kim, N., Lim, W., Park, Y.-I., Cha, S. W., & Jang,
M. E. (2017, October). Optimization of power man-
agement among an engine, battery and ultra-capacitor
for a series HEV: A dynamic programming application.
International Journal of Automotive Technology,18(5),
891–900. doi: 10.1007/s12239-017-0087-4
Juesas, P., Ramasso, E., Drujont, S., & Placet, V. (2016). On
partially supervised learning and inference in dynamic
Bayesian networks for prognostics with uncertain fac-
tual evidence: Illustration with Markov switching mod-
els. , 10.
Jung, W., & Ismail, A. (2015). Prognostic and Health Man-
agement Trend in Automotive Industry: An Overview.
, 7.
Kan, M. S., Tan, A. C., & Mathew, J. (2015, Octo-
ber). A review on prognostic techniques for non-
stationary and non-linear rotating systems. Mechani-
cal Systems and Signal Processing,62-63, 1–20. doi:
10.1016/j.ymssp.2015.02.016
Klabfleisch, J. D., & Prentice, R. L. (2002). The statistical
analysis of failure time data, 2nd edition. New York,
USA: Wiley.
Ko, T., Karayel, D., Boru, B., Ayhan, V., Cesur, ., & Parlak,
A. (2014, January). Design and Implementation of the
Control System of an Internal Combustion Engine Test
Unit. Advances in Mechanical Engineering,6, 914876.
doi: 10.1155/2014/914876
Lasheras, F., Nieto, P., de Cos Juez, F., Bayn, R., & Surez, V.
(2015, March). A Hybrid PCA-CART-MARS-Based
Prognostic Approach of the Remaining Useful Life for
Aircraft Engines. Sensors,15(3), 7062–7083. doi:
10.3390/s150307062
Lee, J., Wu, F., Zhao, W., Ghaffari, M., Liao, L., &
Siegel, D. (2014, January). Prognostics and health
management design for rotary machinery systemsRe-
views, methodology and applications. Mechanical Sys-
tems and Signal Processing,42(1-2), 314–334. doi:
10.1016/j.ymssp.2013.06.004
Li, R., Verhagen, W. J. C., & Curran, R. (2018). A Func-
tional Architecture of Prognostics and Health Manage-
ment using a Systems Engineering Approach. , 10.
Li, S., Zhang, G., & Wang, J. (2017, June). Civil air-
craft health management research based on big data
and deep learning technologies. In 2017 IEEE Inter-
national Conference on Prognostics and Health Man-
agement (ICPHM) (pp. 154–159). Dallas, TX, USA:
IEEE. doi: 10.1109/ICPHM.2017.7998321
Li, X., Ding, Q., & Sun, J.-Q. (2018, April). Remaining use-
ful life estimation in prognostics using deep convolu-
tion neural networks. Reliability Engineering & System
Safety,172, 1–11. doi: 10.1016/j.ress.2017.11.021
Lin, W.-C., & Ghoneim, Y. A. (2016, June). Model-based
fault diagnosis and prognosis for Electric Power Steer-
ing systems. In 2016 IEEE International Conference
on Prognostics and Health Management (ICPHM)
(pp. 1–8). Ottawa, ON, Canada: IEEE. doi:
10.1109/ICPHM.2016.7542840
Lipton, Z. C. (2015). A Critical Review of Recurrent Neural
Networks for Sequence Learning. , 34.
Lipu, M. H., Hannan, M., Hussain, A., Hoque, M., Ker,
P. J., Saad, M., & Ayob, A. (2018, December). A
review of state of health and remaining useful life
estimation methods for lithium-ion battery in electric
vehicles: Challenges and recommendations. Jour-
nal of Cleaner Production,205, 115–133. doi:
10.1016/j.jclepro.2018.09.065
Liu, J., Wang, W., Ma, F., Yang, Y., & Yang, C. (2012). A
data-model-fusion prognostic framework for dynamic
system state forecasting. Engineering Applications of
Artificial Intelligence,25(4), 814 - 823. (Special Sec-
tion: Dependable System Modelling and Analysis) doi:
https://doi.org/10.1016/j.engappai.2012.02.015
Liu, W., Tang, B., Han, J., Lu, X., Hu, N., & He, Z. (2015,
April). The structure healthy condition monitoring and
fault diagnosis methods in wind turbines: A review.
Renewable and Sustainable Energy Reviews,44, 466–
472. doi: 10.1016/j.rser.2014.12.005
Luo, H., Huang, M., & Zhou, Z. (2018). Integration of multi-
gaussian fitting and lstm neural networks for health
31
INT ERNATIO NAL JO UR NAL O F PROGNOSTICS AND HEALT H MANAGEMENT
monitoring of an automotive suspension component.
Journal of Sound and Vibration,428, 87 - 103. doi:
https://doi.org/10.1016/j.jsv.2018.05.007
Luo, J., Namburu, M., Pattipati, K. R., Qiao, L., & Chi-
gusa, S. (2010, March). Integrated Model-Based and
Data-Driven Diagnosis of Automotive Antilock Brak-
ing Systems. IEEE Transactions on Systems, Man,
and Cybernetics - Part A: Systems and Humans,40(2),
321–336. doi: 10.1109/TSMCA.2009.2034481
Luo, J., Pattipati, K. R., Qiao, L., & Chigusa, S. (2008, Sept).
Model-based prognostic techniques applied to a sus-
pension system. IEEE Transactions on Systems, Man,
and Cybernetics - Part A: Systems and Humans,38(5),
1156-1168. doi: 10.1109/TSMCA.2008.2001055
Ma, M. J., Lu, C., Zerhouni, M. N., & Cheng, Y. (2018). Air-
craft Engine Health State Classification Using Stacked
Denoising Autoencoder. , 6.
Matsuishi, M., & Endo, T. (1968). Fatigue of metals sub-
jected to varying stress. Japan Soc. Mech. Engineer-
ing.
Meinhold, R. J., & Singpurwalla, N. D. (1983). Understand-
ing the Kalman Filter.
Meneghetti, L., Terzi, M., Del Favero, S., Susto, G. A., &
Cobelli, C. (2018). Data-Driven Anomaly Recognition
for Unsupervised Model-Free Fault Detection in Arti-
ficial Pancreas. IEEE Transactions on Control Systems
Technology, 1–15. doi: 10.1109/TCST.2018.2885963
Miner, M. (1945). Cumulative damage in fatigue. Journal of
Applied Mechanics,12, A159-A164.
Ming Yu, & Danwei Wang. (2014, July). Model-Based
Health Monitoring for a Vehicle Steering System With
Multiple Faults of Unknown Types. IEEE Transac-
tions on Industrial Electronics,61(7), 3574–3586. doi:
10.1109/TIE.2013.2281159
Naderi, E., Meskin, N., & Khorasani, K. (2011). Nonlin-
ear Fault Diagnosis of Jet Engines by Using a Multiple
Model-Based Approach. , 13.
Naderi, E., Meskin, N., & Khorasani, K. (2012). Nonlin-
ear Fault Diagnosis of Jet Engines by Using a Multi-
ple Model-Based Approach. Journal of Engineering
for Gas Turbines and Power,134(1), 011602. doi:
10.1115/1.4004152
Nguyen, D. V., Limmer, S., Yang, K., Olhofer, M., & B¨
ack, T.
(2019). Modelling and prediction of remaining useful
lifetime for maintenance scheduling optimization of a
car fleet. International Journal of Performability Engi-
neering.
Ordez, C., Snchez Lasheras, F., Roca-Pardias, J., & Juez,
F. J. d. C. (2019, January). A hybrid ARI-
MASVM model for the study of the remaining use-
ful life of aircraft engines. Journal of Computa-
tional and Applied Mathematics,346, 184–191. doi:
10.1016/j.cam.2018.07.008
Paul, S., Kapoor, K., Jasani, D., Dudhwewala, R., Gowda,
V. B., & Nair, T. R. G. (2008). Application of Artificial
Neural Networks in Aircraft Maintenance, Repair and
Overhaul Solutions. , 7.
Paul A. Gagniuc. (2017). Markov Chains: From Theory to
Implementation and Experimentation. USA, NJ: John
Wiley & Sons.
Pecht, M., & Jaai, R. (2010, March). A prognostics and health
management roadmap for information and electronics-
rich systems. Microelectronics Reliability,50(3), 317–
323. doi: 10.1016/j.microrel.2010.01.006
Pecht, M., & Jie Gu. (2009, June). Physics-of-failure-based
prognostics for electronic products. Transactions of the
Institute of Measurement and Control,31(3-4), 309–
322. doi: 10.1177/0142331208092031
Poritz, A. (1982). Linear predictive hidden Markov mod-
els and the speech signal. In ICASSP ’82. IEEE In-
ternational Conference on Acoustics, Speech, and Sig-
nal Processing (Vol. 7, pp. 1291–1294). Paris, France:
Institute of Electrical and Electronics Engineers. doi:
10.1109/ICASSP.1982.1171633
Rabiner, L. R. (1990). Readings in speech recognition. In
A. Waibel & K.-F. Lee (Eds.), (pp. 267–296).
Radac, M., Precup, R., & Petriu, E. M. (2015, Nov). Model-
free primitive-based iterative learning control approach
to trajectory tracking of mimo systems with experi-
mental validation. IEEE Transactions on Neural Net-
works and Learning Systems,26(11), 2925-2938. doi:
10.1109/TNNLS.2015.2460258
Radac, M.-B., & Precup, R.-E. (2017, January). Data-driven
model-free slip control of anti-lock braking systems us-
ing reinforcement Q-learning. Neurocomputing,275,
317–329. doi: 10.1016/j.neucom.2017.08.036
Radac, M.-B., Precup, R.-E., & Roman, R.-C. (2018,
February). Data-driven model reference control of
MIMO vertical tank systems with model-free VRFT
and Q-Learning. ISA Transactions,73, 227–238. doi:
10.1016/j.isatra.2018.01.014
Ramasso, E. (2014). Investigating computational geometry
for failure prognostics. , 18.
Ramasso, E., & Saxena, A. (2014). Performance Benchmark-
ing and Analysis of Prognostic Methods for CMAPSS
Datasets. , 15.
Razavi-Far, R., Chakrabarti, S., Saif, M., & Zio, E. (2019,
January). An integrated imputation-prediction scheme
for prognostics of battery data with missing observa-
tions. Expert Systems with Applications,115, 709–723.
doi: 10.1016/j.eswa.2018.08.033
Rezvanizaniani, S. M., Liu, Z., Chen, Y., & Lee, J.
(2014, June). Review and recent advances in
battery health monitoring and prognostics technolo-
gies for electric vehicle (EV) safety and mobility.
Journal of Power Sources,256, 110–124. doi:
10.1016/j.jpowsour.2014.01.085
Ross, T. J. (2010). Fuzzy logic with engineering applications
32
INT ERNATIO NAL JO UR NAL O F PROGNOSTICS AND HEALT H MANAGEMENT
(3rd ed ed.). Chichester, U.K: John Wiley. (OCLC:
ocn430736639)
Saimurugan, M., Praveenkumar, T., Sabhrish, B.,
Sachin Menon, P., & Sanjiv, S. (2016, Septem-
ber). On-Road Testing of A Vehicle for Gearbox
Fault Detection using Vibration Signals. Indian
Journal of Science and Technology,9(34). doi:
10.17485/ijst/2016/v9i34/100957
Sankavaram, C., Kodali, A., Pattipati, K., Singh, S., Zhang,
Y., & Salman, M. (2016). An Inference-based Prognos-
tic Framework for Health Management of Automotive
Systems. , 16.
Sankavaram, C., Pattipati, B., Kodali, A., Pattipati, K.,
Azam, M., Kumar, S., & Pecht, M. (2009, Au-
gust). Model-based and data-driven prognosis of au-
tomotive and electronic systems. In 2009 IEEE Inter-
national Conference on Automation Science and Engi-
neering (pp. 96–101). Bangalore, India: IEEE. doi:
10.1109/COASE.2009.5234108
Saxena, A., Celaya, J., Balaban, E., Goebel, K., Saha, B.,
Saha, S., & Schwabacher, M. (2008, October). Metrics
for evaluating performance of prognostic techniques.
In 2008 International Conference on Prognostics and
Health Management (pp. 1–17). Denver, CO, USA:
IEEE. doi: 10.1109/PHM.2008.4711436
Saxena, A., Celaya, J., Saha, B., Saha, S., & Goebel, K.
(2009, March). Evaluating algorithm performance met-
rics tailored for prognostics. In 2009 IEEE Aerospace
conference (pp. 1–13). Big Sky, MT, USA: IEEE. doi:
10.1109/AERO.2009.4839666
Saxena, A., Celaya, J., Saha, B., Saha, S., & Goebel, K.
(2010, March). Evaluating prognostics performance
for algorithms incorporating uncertainty estimates. In
2010 IEEE Aerospace Conference (pp. 1–11). Big Sky,
MT: IEEE. doi: 10.1109/AERO.2010.5446828
Saxena, A., & Goebel, K. (2008). C-mapss data set. NASA
Ames Prognostics Data Repository.
Selvi, D., Piga, D., & Bemporad, A. (2018, June). To-
wards direct data-driven model-free design of opti-
mal controllers. In 2018 European Control Confer-
ence (ECC) (pp. 2836–2841). Limassol: IEEE. doi:
10.23919/ECC.2018.8550184
Shafi, U., Safi, A., Shahid, A. R., Ziauddin, S., & Saleem,
M. Q. (2018). Vehicle Remote Health Moni-
toring and Prognostic Maintenance System. Jour-
nal of Advanced Transportation,2018, 1–10. doi:
10.1155/2018/8061514
Shao, Y., Liang, J., Gu, F., Chen, Z., & Ball, A. (2011, July).
Fault Prognosis and Diagnosis of an Automotive Rear
Axle Gear Using a RBF-BP Neural Network. Jour-
nal of Physics: Conference Series,305, 012063. doi:
10.1088/1742-6596/305/1/012063
Shih, T. I.-P., & Yang, V. (Eds.). (2014). Turbine aerodynam-
ics, heat transfer, materials, and mechanics (No. 243).
Reston, Va: American Inst. of Aeronautics and Astro-
nautics, Inc. (OCLC: 903312698)
Shufen, Q., & Wanying, Z. (2013). Prognostic and Health
Management System based on Flight Data. , 3.
Si, X.-S., Wang, W., Hu, C.-H., & Zhou, D.-H. (2011,
August). Remaining useful life estimation A review
on the statistical data driven approaches. European
Journal of Operational Research,213(1), 1–14. doi:
10.1016/j.ejor.2010.11.018
Sikorska, J., Hodkiewicz, M., & Ma, L. (2011a).
Prognostic modelling options for remaining useful
life estimation by industry. Mechanical Systems
and Signal Processing,25(5), 1803–1836. doi:
10.1016/j.ymssp.2010.11.018
Sikorska, J., Hodkiewicz, M., & Ma, L. (2011b, July).
Prognostic modelling options for remaining useful
life estimation by industry. Mechanical Systems
and Signal Processing,25(5), 1803–1836. doi:
10.1016/j.ymssp.2010.11.018
Simon, D. (2008, June). A comparison of filter-
ing approaches for aircraft engine health estimation.
Aerospace Science and Technology,12(4), 276–284.
doi: 10.1016/j.ast.2007.06.002
Singh, S., Kodali, A., & Pattipati, K. (2009, Aug). A
factorial hidden markov model (fhmm)-based rea-
soner for diagnosing multiple intermittent faults.
In 2009 ieee international conference on automa-
tion science and engineering (p. 146-151). doi:
10.1109/COASE.2009.5234134
Sobczyk, K., & Spencer, B. (1993). Random fatigue: From
data to theory. San Diego, USA: San Diego, CA: Aca-
demic.
Spall, J. C., & Cristion, J. A. (1998, Sep.). Model-free con-
trol of nonlinear stochastic systems with discrete-time
measurements. IEEE Transactions on Automatic Con-
trol,43(9), 1198-1210. doi: 10.1109/9.718605
Stainier, L., Leygue, A., & Ortiz, M. (2019, August). Model-
Free Data-Driven Methods in Mechanics: Material
Data Identification and Solvers. Computational Me-
chanics,64(2), 381–393. (arXiv: 1903.07983) doi:
10.1007/s00466-019-01731-1
Su, S., Zhang, W., & Zhao, S. (2014). Fault Prediction
for Nonlinear System Using Sliding ARMA Combined
with Online LS-SVR. Mathematical Problems in En-
gineering,2014, 1–9. doi: 10.1155/2014/692848
Sutharssan, T., Stoyanov, S., Bailey, C., & Yin, C. (2015,
July). Prognostic and health management for engineer-
ing systems: a review of the data-driven approach and
algorithms. The Journal of Engineering,2015(7), 215–
222. doi: 10.1049/joe.2014.0303
Tahan, M., Tsoutsanis, E., Muhammad, M., & Abdul Karim,
Z. (2017, July). Performance-based health monitoring,
diagnostics and prognostics for condition-based main-
tenance of gas turbines: A review. Applied Energy,
33
INT ERNATIO NAL JO UR NAL O F PROGNOSTICS AND HEALT H MANAGEMENT
198, 122–144. doi: 10.1016/j.apenergy.2017.04.048
Taie, M. A., Diab, M., & ElHelw, M. (2012, Oc-
tober). Remote prognosis, diagnosis and main-
tenance for automotive architecture based on least
squares support vector machine and multiple clas-
sifiers. In 2012 IV International Congress on Ul-
tra Modern Telecommunications and Control Systems
(pp. 128–134). St. Petersburg, Russia: IEEE. doi:
10.1109/ICUMT.2012.6459652
Tan, K. K., Lee, T. H., Huang, S. N., & Leu, F. M. (2001,
Apr 01). Adaptive-predictive control of a class of siso
nonlinear systems. Dynamics and Control,11(2), 151–
174. doi: 10.1023/A:1012583811904
Tang, L., Kacprzynski, G. J., Goebel, K., Saxena, A., Saha,
B., & Vachtsevanos, G. (2008, Oct). Prognostics-
enhanced automated contingency management for ad-
vanced autonomous systems. In 2008 international
conference on prognostics and health management
(p. 1-9). doi: 10.1109/PHM.2008.4711448
Tian, J., Xiong, R., & Yu, Q. (2019, February). Fractional-
Order Model-Based Incremental Capacity Analysis for
Degradation State Recognition of Lithium-Ion Bat-
teries. IEEE Transactions on Industrial Electronics,
66(2), 1576–1584. doi: 10.1109/TIE.2018.2798606
Tinga, T. (2013). Springer Series in Reliability Engineering.
London, Heidelberg, New York, Dordrecht: Springer.
Tipping, M. E. (1999). The Relevance Vector Machine. , 7.
Tsui, K. L., Chen, N., Zhou, Q., Hai, Y., & Wang, W. (2015).
Prognostics and Health Management: A Review on
Data Driven Approaches. Mathematical Problems in
Engineering,2015, 1–17. doi: 10.1155/2015/793161
Vachtsevanos, G., Lewis, F., Roemer, M., Hess, A., & Wu,
B. (2006). Intelligent Fault Diagnosis and Progno-
sis for Engineering Systems: Vachtsevanos/Intelligent
Fault Diagnosis. Hoboken, NJ, USA: John Wiley &
Sons, Inc. doi: 10.1002/9780470117842
Vapnik, V. N. (1995). The Nature of Statistical Learning
Theory. New York, NY: Springer New York. (OCLC:
905485685)
Vatani, A., Khorasani, K., & Meskin, N. (2015, June).
Health Monitoring and Degradation Prognostics in
Gas Turbine Engines Using Dynamic Neural Net-
works. In Volume 6: Ceramics; Controls, Diag-
nostics and Instrumentation; Education; Manufactur-
ing Materials and Metallurgy; Honors and Awards
(p. V006T05A030). Montreal, Quebec, Canada:
ASME. doi: 10.1115/GT2015-44101
Vianna, W. O. L., Rodrigues, L. R., & Yoneyama, T. (2015).
Aircraft Line Maintenance Planning Based on PHM
Data and Resources Availability Using Large Neigh-
borhood Search. , 7.
Vogl, G. W., Weiss, B. A., & Donmez, M. A. (2014). Stan-
dards for Prognostics and Health Management (PHM)
Techniques within Manufacturing Operations. , 13.
Vogl, G. W., Weiss, B. A., & Helu, M. (2016, Jun 09). A
review of diagnostic and prognostic capabilities and
best practices for manufacturing. Journal of Intelligent
Manufacturing. doi: 10.1007/s10845-016-1228-8
Wang, D., Miao, Q., & Pecht, M. (2013). Prognos-
tics of lithium-ion batteries based on relevance vectors
and a conditional three-parameter capacity degradation
model. Journal of Power Sources,239, 253 - 264. doi:
https://doi.org/10.1016/j.jpowsour.2013.03.129
Wang, J., Mao, X., Zhu, K., Song, J., & Zhuo, B. (2009,
September). An intelligent diagnostic tool for electron-
ically controlled diesel engine. Mechatronics,19(6),
859–867. doi: 10.1016/j.mechatronics.2009.04.009
Wang, M.-H., Chao, K.-H., Sung, W.-T., & Huang, G.-J.
(2010, April). Using ENN-1 for fault recognition of
automotive engine. Expert Systems with Applications,
37(4), 2943–2947. doi: 10.1016/j.eswa.2009.09.041
Wang, T., Jianbo Yu, Siegel, D., & Lee, J. (2008, October). A
similarity-based prognostics approach for Remaining
Useful Life estimation of engineered systems. In 2008
International Conference on Prognostics and Health
Management (pp. 1–6). Denver, CO, USA: IEEE. doi:
10.1109/PHM.2008.4711421
Wang, Y., Limmer, S., Olhofer, M., Emmerich, M. T. M.,
& B¨
ack, T. (2019, June). Vehicle fleet main-
tenance scheduling optimization by multi-objective
evolutionary algorithms. In 2019 ieee congress on
evolutionary computation (cec) (p. 442-449). doi:
10.1109/CEC.2019.8790142
Wheeler, K. R., Kurtoglu, T., & Poll, S. D. (2009). A Sur-
vey of Health Management User Objectives Related
to Diagnostic and Prognostic Metrics. In Volume 2:
29th Computers and Information in Engineering Con-
ference, Parts A and B (pp. 1287–1298). San Diego,
California, USA: ASME. doi: 10.1115/DETC2009-
87073
Wu, J. (2017). Introduction to Convolutional Neural Net-
works. , 31.
Xia, T., Dong, Y., Xiao, L., Du, S., Pan, E., & Xi,
L. (2018). Recent advances in prognostics and
health management for advanced manufacturing
paradigms. Reliability Engineering and Sys-
tem Safety,178(C), 255-268. Retrieved from
https://ideas.repec.org/a/eee/reensy/v178y2018icp255-268.html
doi: 10.1016/j.ress.2018.06.02
Xing, Y., Ma, E. W., Tsui, K.-L., & Pecht, M. (2013).
An ensemble model for predicting the remaining
useful performance of lithium-ion batteries. Mi-
croelectronics Reliability,53(6), 811 - 820. doi:
https://doi.org/10.1016/j.microrel.2012.12.003
Xu, J., Wang, Y., & Xu, L. (2014, April). PHM-Oriented In-
tegrated Fusion Prognostics for Aircraft Engines Based
on Sensor Data. IEEE Sensors Journal,14(4), 1124–
1132. doi: 10.1109/JSEN.2013.2293517
34
INT ERNATIO NAL JO UR NAL O F PROGNOSTICS AND HEALT H MANAGEMENT
Xu, J., & Xu, L. (2011, June). Health management based
on fusion prognostics for avionics systems. Journal of
Systems Engineering and Electronics,22(3), 428–436.
doi: 10.3969/j.issn.1004-4132.2011.03.010
Yang, C., Song, P., & Liu, X. (2017, Apr 18). Failure
prognostics of heavy vehicle hydro-pneumatic spring
based on novel degradation feature and support vector
regression. Neural Computing and Applications. doi:
10.1007/s00521-017-2986-8
Yang, F., Xing, Y., Wang, D., & Tsui, K.-L. (2016,
February). A comparative study of three model-based
algorithms for estimating state-of-charge of lithium-
ion batteries under a new combined dynamic load-
ing profile. Applied Energy,164, 387–399. doi:
10.1016/j.apenergy.2015.11.072
Yang, L., Wang, J., & Zhang, G. (2016, June). Aviation
PHM system research framework based on PHM big
data center. In 2016 IEEE International Conference
on Prognostics and Health Management (ICPHM)
(pp. 1–5). Ottawa, ON, Canada: IEEE. doi:
10.1109/ICPHM.2016.7542824
Ye, M., Guo, H., & Cao, B. (2017, March). A
model-based adaptive state of charge estimator for a
lithium-ion battery using an improved adaptive par-
ticle filter. Applied Energy,190, 740–748. doi:
10.1016/j.apenergy.2016.12.133
You, G.-w., Park, S., & Oh, D. (2016, August). Real-time
state-of-health estimation for electric vehicle batteries:
A data-driven approach. Applied Energy,176, 92–103.
doi: 10.1016/j.apenergy.2016.05.051
Zhang, J., & Lee, J. (2011, August). A review on prog-
nostics and health monitoring of Li-ion battery. Jour-
nal of Power Sources,196(15), 6007–6014. doi:
10.1016/j.jpowsour.2011.03.101
Zhang, M. Y., Liu, J., Hanachi, D. H., Yu, M. X., & Yang,
M. Y.-B. (2018). Physics-based Model and Neural
Network Model for Monitoring Starter Degradation of
APU. , 7.
Zhang, X., Kang, J., Zhao, J. S., & Cao, D. C. (2013, July).
Features for fault diagnosis and prognosis of gearbox.
Chemical Engineering Transactions, 1027–1032. doi:
10.3303/CET1333172
Zhao, Z., Liang, B., Wang, X., & Lu, W. (2017). Re-
maining useful life prediction of aircraft engine based
on degradation pattern learning. Reliability Engi-
neering and System Safety,164, 74 - 83. doi:
https://doi.org/10.1016/j.ress.2017.02.007
Zheng, X., & Fang, H. (2015). An integrated un-
scented kalman filter and relevance vector regres-
sion approach for lithium-ion battery remaining use-
ful life and short-term capacity prediction. Reliabil-
ity Engineering & System Safety,144, 74 - 82. doi:
https://doi.org/10.1016/j.ress.2015.07.013
Zhong, S., Li, M. Z., Lin, M. L., & Zhang, D. Y.
(2018). Aero-Engine Exhaust Gas Temperature Prog-
nostic Model Based on Gated Recurrent Unit Network.
, 5.
Zou, K.-X., Ma, H.-D., Fang, H.-Z., & Yi, D.-W. (2011,
May). Study of prognostics for spacecraft based-on
particle swarm optimized neural network. In 2011
Prognostics and System Health Managment Confer-
nece (pp. 1–5). Shenzhen, China: IEEE. doi:
10.1109/PHM.2011.5939479
35
