Some Challenges and Opportunities in Reliability

Engineering
Enrico Zio

To cite this version:

Enrico Zio. Some Challenges and Opportunities in Reliability Engineering. IEEE Transactions on
Reliability, 2016, 65 (4), pp.1769-1782. �10.1109/TR.2016.2591504�. �hal-01550063�

HAL Id: hal-01550063

https://hal.science/hal-01550063
Submitted on 29 Jun 2017

HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est

archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents
entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non,
lished or not. The documents may come from émanant des établissements d’enseignement et de
teaching and research institutions in France or recherche français ou étrangers, des laboratoires
abroad, or from public or private research centers. publics ou privés.
 Some challenges and opportunities in reliability engineering

Enrico Zioa,b
a
Chair System Science and the Energy Challenge, Fondation Electricite’ de France (EDF), CentraleSupélec, Université

Paris-Saclay, Grande Voie des Vignes, 92290 Chatenay-Malabry, France

b
Dipartimento di Energia - Politecnico di Milano

Via La Masa 34, I-20156 Milano, Italy

Abstract

Today’s fast-pace evolving and digitalizing World is posing new challenges to reliability
engineering. On the other hand, the continuous advancement of technical knowledge and the
increasing capabilities of monitoring and computing offer opportunities for new developments in
reliability engineering.

In this paper, I reflect on some of these challenges and opportunities in research and application. The
underlying perspective taken stands on:

 the belief that the knowledge, information and data (KID) available for the modeling,
computations and analyses done in reliability engineering is substantially grown and continue
to do so;
 the belief that the technical capabilities for reliability engineering have been significantly
advanced;
 the recognition of the increased complexity of the systems, nowadays more and more made
of heterogeneous, highly interconnected elements.

In line with this perspective, opportunities and challenges for reliability engineering are discussed in
relation to degradation modeling and integration of multi-state and physics-based models therein,
accelerated degradation testing, component-, system- and fleet-wide prognostics and health
management in evolving environments.

The paper is not a review, nor a state of the art work, but rather it offers a vision of reflection on
reliability engineering, for consideration and discussion by the interested scientific community. It
does not pretend to give the unique view, nor to be complete in the subject discussed and the related
literature referenced to.
Keywords: Degradation Modeling, Physics-Based Models, Dependent Degradation, Random Shocks,
Multi-State System Reliability, Piecewise Deterministic Markov Process, Accelerated Degradation
Testing, Prognostics and Health Management, Prognostic Performance Indicator, Return of
Investment, Evolving Environment, Fleet Prognostics, Distributed Prognostics

List of acronyms

AAKR Auto-Associative Kernel Regression

ADT Accelerated Degradation Testing

AIC Akaike’s Information Criterion

ALT Accelerated Life Testing

ANN Artificial Neural Network

BBD Binary Decision Diagrams

BMA Bayesian Model Averaging

CBM Condition-Based Maintenance

CL Cost of Loss

CM Corrective Maintenance

CMS Condition Monitoring System

D-S theory Dempster-Shafer theory

EE Evolving Environment

FA Failures Avoided

HI Health Index

KID Knowledge, Information and Data

KNN K-Nearest Neighbors

LED Light Emitting Diode

MOSFET Metal Oxide Semiconductor Field Effect Transistor

MSM Multi-State Model

 MSPM Multi-State Physics Model

PBM Physics-Based Model

PDMP Piecewise Deterministic Markov Processes

PHM Prognostics and Health Management

PI Precision Index

PrM Preventive Maintenance

PMS Phased-Mission Systems

PPI Prognostic Performance Indicator

RI Risk Index

ROI Return On Investment

RUL Remaining Useful Life

RVM Relevance Vector Machines

SPRT Sequential Probability Ratio Test

SVM Support Vector Machines

1. Introduction

As the digital, physical and human worlds continue to integrate, we experience a deep
transformation in industry, which far-reaches into our lives. The 4th industrial revolution, the
internet of things and big data, the industrial internet, are changing the way we design,
manufacture, provide products and services. This is creating a complex network of things and
people that are seamlessly connected and communicating. It is providing opportunities to make
productions systems more efficient and faster, and more flexible and resilient the complex supply
chains and distribution networks that tie the global economy.

In this fast-pace changing environment, the attributes related to the reliability of components and
systems continue to play a fundamental role for industry. The innovations that are being
developed have high potential of increased wellbeing and benefits, but also generate new and
unknown failure mechanisms, new and unknown functional and structural dependencies, and
eventually new and unknown hazards and risks. On the other hand, the advancements in
knowledge, methods and techniques, the increase in information sharing and data availability,
offer new opportunities of analysis and assessment for reliability engineering. Then, a new
reliability engineering “revolution” is in the making for addressing the challenges brought about
by the new and evolved systems, and the innovations therein; this calls for and, at the same time,
drives the advancements of new methods and tools, and the extension of their applications, based
on the increased knowledge, information and data (KID) available, which can improve our
reliability prediction capability.

In this paper, I consider the above context and address some challenges and opportunities for
reliability engineering. In particular, I argue that the big KID available allows refined modeling
of the degradation processes that drive the components and systems to failure.

In Section 2, multi-state modeling schemes are discussed as an opportunity of describing the

degradation processes as Piecewise Deterministic Markov Processes (PDMP), accommodating
the relevant knowledge in physics-based models of transition rates (Multi-State Physics-based
Models, MSPM). Challenges are highlighted in relation to the treatment of dependent, and often
competing, processes, the accounting of uncertainties and of the effects of maintenance. On the
contrary, the complexity of the resulting models does not impede their solution by advanced
mathematical techniques and Monte Carlo simulation, although scalability may be a challenge
for large systems. As for the problem of the estimation of the model parameters values, this can
be tackled by Accelerated Degradation Testing (ADT), as discussed in the subsequent Section 3.

On the other hand, the increased availability of data coming from monitoring the relevant
components and systems parameters and the grown ability of treating these data by intelligent
 algorithms capable of mining out information relevant to the assessment and prediction of their
state, has open wide the doors for Prognostics and Health Management (PHM) in many industrial
sectors, for improved operation and maintenance. This is discussed in Section 4, highlighting the
challenges of:

 developing methods and models capable of dealing with the inevitable Evolving
Environment (EE) that changes the conditions in which a component or system lives, and
degradation and failure occur (PHM in EE);
 considering the information and data coming from a fleet of components and systems, for
effectively informed PHM (Fleet PHM);
 considering the integration and interaction of the information and data monitored on
multiple components operating on a system (Distributed PHM);
 evaluating the performance of PHM approaches by defining significant Prognostic
Performance Indicators (PPIs) and their adequacy for a given application, by a maturity
assessment of the predictive model;
 evaluating the economic and safety value of a PHM approach for condition-based and/or
predictive maintenance.

2. Degradation modeling

In the evolving scenario of development of complex networks of things and people, supported by
(more and more interdependent) critical infrastructures, concerns are arising on vulnerability to
failure and risk of accident. The worries are that the allocated system capacities may not be adequate
to respond to the postulated growing demands and that the safety margins preventively designed may
degrade and become insufficient to cope with the expected and unexpected stresses arriving onto the
systems. Emergent behaviors may arise in collective ways difficult to predict from the superposition
of the behavior of the individual elements and difficult to manage resiliently. Indeed, we are
witnessing more and more system-level breakdowns in our critical infrastructures, which emerge
from local degradations of individual components that cascade to large-scale consequences possibly
propagating from one system to another, through the (inter-)dependencies. The analysis of this
problem for ensuring the protection and resilience of our critical infrastructures calls for the
integration of methods capable of viewing the problem from different perspectives (topological and
functional, static and dynamic, etc.), under the existing uncertainties and given the high system
complexity (Kröger and Zio 2011).

On the other hand, safety-critical components and systems, like those employed in the nuclear, oil
and gas, automotive, aeronautic and aerospace sectors, are designed not to fail, i.e. with very high
reliability, because of the potentially catastrophic consequences of their failures. Traditional data-
based reliability analysis, based on failure data, is, then, unsuitable. Yet, most failure mechanisms
can be traced to underlying degradation processes (e.g. wear, stress corrosion, shocks, cracking,
fatigue, etc.), for which models may exist (Yang, 2002).

In general, the reliability of a system decreases as the degradation processes develop, eventually
leading to failure (Ye and Xie, 2015). In reliability engineering, degradation processes have been
widely studied and different degradation models have been developed.

However, there is a need for the development of a holistic framework of models and computational
methods for the reliability-based analysis and maintenance optimization of safety-critical systems,
taking into account the available knowledge about the components and systems degradation and
failure behaviors, their dependencies, the external influencing factors and the associated
uncertainties. The holistic treatment of the problem should also allow for the consideration of
damage precursor, as backbones for tracking component and system degradation which can allow
early detection and effective, predictive maintenance.

The existing degradation models are often classified into the following general categories:

 statistical models of time to failure, based on degradation data (e.g. Bernstein distribution
(Gebrraeel et al., 2009), Weibull distribution (Lu and Meeker, 1993));

 stochastic process models (e.g. Gamma process (Lawless and Crowder, 2004), inverse
Gaussian process (Chen et al., 2015)) describing the evolution of one or more degradation
parameters by gradual, stochastic degradation increments over time, and the failure occurs
when the degradation parameter values reach predefined thresholds;

 physics-based models (PBMs), based on the knowledge of the physics of degradation, which
is translated into equations to give a quantitative description (e.g. the physics functions based
on critical environmental stresses, e.g. amplitude and frequency of mechanical loads, used to
model the pitting and corrosion-fatigue degradation mechanisms (Chookah et al, 2011));

 multi-state models (MSMs) describing the underlying degradation process by finite

degradation states (e.g. semi-Markov models for the deterioration of infrastructure systems
(Black et al., 2005); piecewise deterministic Markov process for dependent degradation
process (Lin et al., 2014)).

The recent literature on degradation modeling can be organized under the above taxonomy. For
statistical models, Lu et al. (Lu et al., 1997) have combined random regression coefficients and a
standard deviation function for analyzing linear degradation data for statistical inference of a time-to-
failure distribution. Lu and Meeker (Lu and Meeker, 1993) have developed methods using
degradation measures to estimate a time-to-failure distribution for a broad class of degradation
models and demonstrated some special cases for which it is possible to obtain closed-form
expressions of the distributions. Yang and Yang (Yang and Yang, 1998) have estimated the
parameters of lifetime distributions using a random-coefficient-based approach that uses the lifetimes
of failed devices, combined with degradation information from operating devices.

For stochastic models, Whitmore (Whitemore, 1995) has estimated the degradation process by a
Wiener diffusion process subject to measurement errors due to imperfect instruments, procedures and
environments. Lawless and Crowder (Lawless and Crowder, 2004) have constructed a tractable
Gamma-process model incorporating a random effect for taking into account different degradation
rates of the individual components. Chen et al. (Chen et al., 2015) have employed the inverse
Gaussian process with random-drift mode, in which the random drifts are used to represent
heterogeneities commonly observed across the product population. Note that the aforementioned
degradation models are always built on sufficient degradation/failure data.

Physics-Based Models (PBMs) (Daigle and Goebel, 2011; Reggiani et al., 2011; Keedy and Feng,
2012) and Multi-State Models (MSMs) (Moghaddass and Zuo, 2014; Lisnianski and Levitin, 2003;
Li and Pham, 2005; Lin et al., 2012) can be used to describe the evolution of degradation in
structures, systems and components, for which statistical degradation/failure data are insufficient,
e.g. the highly reliable devices in the nuclear, oil and gas, automotive, aeronautic and aerospace
industries. For PBMs, Daigle and Goebel (Daigle and Goebel, 2011) have developed a physics model
of a pneumatic valve, based on mass and energy balances in which the damages depend on sliding
velocity. Reggiani et al. (Reggiani et al., 2011) have developed a physics-based analytical expression
of the linear drain current for hot-carrier stress degradation in transistors. Keedy and Feng (Keedy
and Feng, 2012) have proposed a probabilistic reliability and maintenance modeling framework for
stent deployment and operation, based on physics-of-failure mechanisms, e.g. delayed failure due to
fatigue crack and instantaneous failure due to overload fracture.

For MSMs, Moghaddass and Zuo (Moghaddass and Zuo, 2014) have employed the nonhomogeneous
continuous-time hidden semi-Markov process to model the degradation and observation processes
associated with the device. Giorgio et al. (Giorgio et al., 2011) have developed an age- and state-
dependent Markov model for the wear process of cylinder liners of identical heavy-duty diesel
engines for marine propulsion. Unwin et al. (Unwin et al., 2011) have proposed a multi-state physics
model (MSPM) for the cracking process of a dissimilar metal weld in a primary coolant system of a
nuclear power plant.

Yet, there are several factors which can influence degradation evolution and, thus, need to be
accounted for in degradation modeling.
Indeed, in practice components and systems are often subject to multiple competing degradation
processes and any of them may cause failure (Wang and Pham, 2012). The dependencies among
these processes within one component (e.g. the wear of rubbing surfaces influenced by the
environmental stress shock within a micro-engine (Jiang et al., 2012)), or/and among different
components (e.g. the degradation of the pre-filtrations stations leading to a lower performance level
of the sand filter in a water treatment plant (Rasmekomen and Parlikad, 2013)) need to be
considered, under certain circumstances. Components can be dependent due to functional
dependence, where the failure of a trigger component causes other components to become
inaccessible or unusable (Xing et al., 2012; Wang et al., 2012). Failure isolation effects can induce
degradation dependency among different components, since failure of one component may cause
other components within the same system to become isolated from the system due to the failure
isolation actions (Xing and Levitin, 2010; Wang et al., 2013). This renders challenging the analysis
and prediction of the components and systems reliability (Peng et al., 2010). Wang and Pham (Wang
and Pham, 2012) applied time-varying copulas for describing the dependencies between the
degradation processes modeled by statistical distributions. Straub (Straub, 2009) used a dynamic
Bayesian network to represent the dependencies between degradation processes modeled by multi-
state models. However, very few studies have considered degradation dependency in a system whose
degradation processes are modeled by PBMs and MSMs.

Components may also suddenly fail due to randomly occurring events of excessive loading or
environmental conditions (e.g. excessive temperature) (Wang et al., 2011). For example, thermal and
mechanical shocks (e.g. internal thermal shocks and water hammers) (Lydell, 2000; Salonen et al.,
2007) onto power plant components can lead to intense increases in temperatures and stresses,
respectively. These events, referred to as random shocks, need to be accounted for on top of the
underlying degradation processes, because they can contribute to accelerating the degradation
processes. In the literature, random shocks are typically modeled by Poisson processes (Li and Pham,
2005), distinguishing two main types, extreme shock and cumulative shock processes (Bai et al.,
2006), according to the severity of the damage. The former could directly lead the component to
immediate failure (Anderson, 1987), whereas the latter increases the degree of damage in a
cumulative way (Agrafiotis and Tsoukalas, 1995). Esary and Marshall (1973) have considered
extreme shocks in a component reliability model, whereas Wang et al. (2011), Klutke and Yang
(2002) and Wortman et al. (1994) have modeled the influences of cumulative shocks on a
degradation process. Both extreme and cumulative random shocks have been considered by Li and
Pham (2005), and Wang and Pham (2012). Additionally, Ye et al. (2011) and Fan et al. (2000) have
considered that a high severity of degradation can lead to a high probability that a random shock
causes extreme damage. However, the fact that the effects of cumulative shocks can vary according
to the severity of degradation has also to be considered.
Besides, previous research has focused on the dependency between continuous/multi-state
degradation processes and random shocks. For continuous degradation processes, Peng et al. (2010)
considered systems with one linear degradation path where shocks can bring additional abrupt
degradation damage if the shock loads do not exceed the maximum strength of the material. Multi-
component systems subject to multiple linear degradation paths have been further considered by
Song et al. (2014). Jiang et al. (2012) studied changes in the maximal strength of the material when
systems are deteriorating under different situations. Becker et al. (2002) extended the theory of
dynamic reliability to incorporate random changes of the degradation variables due to random
shocks. Rafiee et al. (2014) proposed reliability models for systems for which the degradation path
has a changing degradation rate according to particular random shock patterns. Song et al. (2014)
studied random shocks with specific sizes or functions, which can selectively affect the degradation
processes of one or more components (not necessarily all components) in one system. For multi-state
degradation processes, Yang et al. (2011) combined random shocks with Markov degradation models
where shocks can lead the systems to further degraded states. However, few studies have explicitly
considered both the dependencies between degradation processes and the random shocks, and among
the degradation processes themselves.

Finally, it is important to consider that the degradation processes can be interrupted or slowed down
by maintenance tasks (e.g. one component can be restored to its initial state by preventive
maintenance if any of its degradations exceed the respective critical level (1992) and by corrective
maintenance upon its failure (2012)). The interactions among components complicate the modeling
for maintenance planning, which becomes a big challenge (2008). Thomas (1986) has categorized
these interactions in maintenance modeling into three groups: economic, structural and stochastic
dependences. Economic dependence exists when the maintenance cost of several components is not
equal to the sum of their individual maintenance costs. For example, Castanier et al. (2005) have
considered a condition-based maintenance policy for a two-unit deteriorating system, where the set-
up cost of inspection is charged only once if the actions on the two components are combined. Van
Dijkhuizen (2000) has investigated the long-term grouping of preventive maintenance jobs in a
multi-setup, multi-component production system where the set-up activities can be combined when
several components are maintained at the same time. Structural dependence occurs if some working
components need to be replaced or dismantled in order to execute the maintenance of the failed ones.
For example, Dekker et al. (1998) have studied the maintenance policy for asphalt roads, where the
number of maintenance services is limited by integrating neighboring segments into a homogeneous
section which is completely repaired. Stochastic dependence, also referred to as probabilistic
dependence, applies when the state of one component can affect those of other components or their
failure rates. Failure interactions have been the most discussed cases for stochastic dependence
(Rasmekomen and Parlikad, 2013) and imply that the failure of one component may lead to the
failure of other components with certain probabilities, and/or influence their failure rates (Murthy
and Nguyen, 1985). For example, Lai and Chen (Lai and Chen, 2006) have presented an economic
periodic replacement model for a two-unit system where the failure of unit 1 can increase the failure
rate of unit 2, while the failure of unit 2 induces unit 1 into instantaneous failure. Zequeira and
Bérenguer (2005)] have studied the inspection policies for a two-component standby system, where
the failure of one component can modify the conditional failure probability of the component still in
operation with probability 𝑝 and does not modify it with probability 1 − 𝑝. Barros et al. (2006) have
optimized the maintenance policy for a two-unit parallel system where the failure of a component
increases the failure rate of the surviving one.

Dependency among degradation mechanisms or processes has received less attention within the
framework of maintenance modeling and optimization of multi-component systems, although they
are of real concern in practice (e.g. the failure of a pump due to oxidation of contacts and bear
wearing). Peng et al. (2010) have developed a maintenance policy with periodic inspections when
two dependent or correlated failure processes are considered. Jiang et al. (2012) have further
compared two preventive maintenance (PrM) policies, age replacement policy and block replacement
policy, combining immediate corrective replacement in consideration of shifting failure thresholds.
Özekici (1988) has considered interdependent aging processes between components due to
continuous wear and shocks, and proposed an optimal periodic replacement policy. Rasmekomen and
Parlikad (2013) have considered degradation dependency in terms of output performance between
one critical component and other parallel components based on aging processes, and the optimal age-
based maintenance policy for this case was also studied. Yang et al. (2013) have proposed a general
statistical reliability model for repairable multi-component systems considering dependent
competing risks, under a partially perfect repair assumption which considers that only the failed
component, rather than the whole system, is replaced. Hong et al. (2014) have used copulas to model
degradation dependency among all the components of a system and obtained the optimal
maintenance policy including condition-based maintenance with periodic inspections and
instantaneous corrective maintenance (CM). Van Horenbeek and Pintelon (2013) have proposed a
dynamic predictive maintenance policy that minimizes the long-term mean maintenance cost per unit
time while considering different component dependencies (i.e. economic, structural and stochastic
dependence). Song et al. (2014) have applied age replacement policy and inspection-based
maintenance policy for systems whose components have s-dependent failure times, and the optimal
replacement interval or inspection times are determined. Note that maintenance optimization for
multi-component systems with multiple dependent competing degradation processes within
individual components has not been considered and only the pre-scheduled periods for inspection or
maintenance are considered as the decision variables of the optimization problem.

From the above description and discussion of the degradation processes, and the associated relevant
factors of dependency, random shocks and maintenance effects, it seems evident that a component-
level and system-level, holistic framework of models and computational methods of reliability
analysis and maintenance optimization is necessary to integrate the available data and knowledge on
degradation processes and failure mechanisms, their dependencies, the external influencing factors
and the associated uncertainties. One main argument in support to this relates to the need of
considering dependencies among different aspects of the components and systems degradation
processes (dependencies among multiple competing degradation processes within one component or
among different components, functional dependencies among components, etc.). It would, indeed, be
very advantageous to have a holistic degradation modelling framework that allows considering in an
integrated manner all such dependencies at component-level and system-level. More specifically, the
availability of such modeling framework would be strongly beneficial for the asset management of
these components and systems, because it would enable to realistically predict component and
system degradation, and optimally plan the necessary maintenance activities. However, it is
important to recognize that at the current state of practice, it is usually not feasible to integrate the
consideration of all these aspects and incorporate them in a single application problem. For
practicality purposes, modeling solutions must be developed, that enable such holistic description of
the degradation processes occurring in a component or system, and their quantitative evaluation by
efficient computation.

Finally, such modelling framework would be apt to deal with phased-mission systems (PMS), such
as those encountered in nuclear, aerospace, chemical, electronic and other industries. These systems
are required to perform different tasks either during any given mission or their operational life time.
For example, a modern aircraft flight typically involves automated take-off, ascent, level flight,
altered flight due to interferences, descent and landing where each of these phases may require
different configurations of components and develop under different environments. In a boiling water
reactor, a loss of coolant accident involves three phases for emergency core cooling: initial core
cooling, suppression core cooling, and residual heat removal. Other systems, such as communication
satellites, may require transportation to the operation site, followed by deployment and a multitude of
on-station activities. Different systems or system configurations may be used during different phases
in different environments, under varying degrees of stress and with consequent different degradation
evolution. In some cases, under certain conditions, a degraded system may be capable of continuing
the phases until mission completion. The reliability analysis of phased-mission systems must, then,
account for changes in configuration, component use, stresses and degradation evolution. Models of
PMS can be built using combinatorial models, Fault Trees, Markov models, Petri nets, Binary
Decision Diagrams (BDDs), simulation models and others (Alam and Al-Saggaf, 1986; Bondavalli et
al., 2004; Dong et al., 2008; Dugan, 1991; Fussel et al., 1981; Lai and Chen, 2006; Laskey, 1996;
Mura and Bondavalli, 1999 and 2001; Sormani et al., 1992; Smotherman and Zermoudeh, 1989;
Wang et al., 2012; Xing and Dugan, 2002 and 2004; Xing and Levitin, 2013; Yong and Dugan, 2004;
Zang et al., 1999). During a particular phase, only certain system features are important, in
accordance with the specific task(s) being performed in that phase. Typically, each phase is identified
by: phase number; duration; system configuration or task(s) to be performed; performance
measure(s) of interest; and maintenance policy. The whole mission is described by a mission profile
table. The overall reliability of a PMS is the probability that the mission successfully achieves (all)
the submission objectives in each phase.

In the context of PMS reliability analysis, the modeling framework discussed above would be quite
useful to capture the relevant factors of dependency, random shocks and maintenance effects,
pertaining to each different phase over the entire mission duration, within an integrated analysis for
the calculation of system and mission performance indicators such as the probability of mission
success, the mean time to failure, the ranks of importance of the system features in the different
phases of the mission.

3. Accelerated Degradation Testing

As mentioned in the previous Section, many components and systems, particularly those employed
in safety-critical applications, are designed to be highly reliable and to have a long lifespan, e.g.
battery life of 15 years for hybrid electric vehicles (Chalk and Miller, 2006).

Traditional reliability tests are obviously not suitable for the reliability assessment of such equipment
over such long time spans. Thus, accelerated degradation tests (ADT) are widely used to accelerate
the failure/degradation processes, exposing the equipment to severe test conditions. Successful
applications of ADT have been developed for batteries (Thomas et al., 2008), light emitting diodes
(LED)s (Wang and Chu, 2012), metal oxide semiconductor field effect transistors (MOSFETs)
(Santini et al., 2014), and others.

In standard ADT data analysis, a degradation model like those discussed in the previous Section is
assumed to describe the degradation paths of the samples tested at different stress levels and some
specific parameters of the model are assumed to be stress-related, as described by a given
acceleration model, e.g. the drift coefficient in the Wiener process (Whitemore and Schenkelberg,
1997; Park and Padgett, 2006; Lim and Yum, 2011). In general, acceleration models are assumed
based on the physical mechanisms of the tested samples or empirical observations of the stress
variable (Escobar and Meeker, 2006), e.g. temperature-Arrhenius model, voltage-Eyring model, etc.
After obtaining the data from ADT, statistical inferences are made to get estimates of the unknown
parameters in both degradation and acceleration models. Then, the component reliability assessment
and lifetime evaluation are performed with the estimated parameters, and considering the given use
conditions. Statistical inference methods for ADT data analysis have been extensively reviewed in
Nelson (1990) and Meeker and Escobar (1998).

In literature, stochastic process models have drawn more attention than degradation-path models,
because of their properties of time-dependent structures, like the Wiener (Whitemore and
Schenkelberg, 1997; Lim and Yum, 2011; Liao and Elsayed, 2006), Gamma (Tseng et al., 2009;
Ling et al., 2015) and inverse Gaussian (Wang and Xu, 2010; Ye and Chen, 2013; Ye et al., 2014;
Peng et al., 2014) process models mentioned in the previous Section.

The Wiener process model is often used when the degradation process is increasing or decreasing
with time. If the decreasing data is discarded, Gamma or inverse Gaussian processes can be used as
degradation models.

Under a unified modeling framework for ADT analysis, it is assumed that the degradation X(t)
follows a stochastic process with statistically independent increments, where the mean and variance
of X(t) are proportional to time. The unified stochastic process thereby defined becomes the Wiener
process model when X(t) follows a normal distribution, a Gamma process when X(t) follows a
Gamma distribution and an inverse Gaussian process when X(t) follows an inverse gamma
distribution.

The acceleration model describes the relationship between the accelerated stress and the degradation
rate. It can be obtained based on either physical knowledge of the tested equipment or empirical
observations. The typical physics-based acceleration models include Arrhenius model, Eyring model,
etc. (Escobar and Meeker, 2006), while the empirical acceleration models include, for example,
Coffin-Manson model (Musallam et al., 2014), etc. A general log-linear form of the model can be
written, whose vector of unknown parameters can be obtained by maximizing the corresponding log-
likelihood functions.

Then, a p−quantile lifetime of interest can derived from the unified stochastic process of accelerated
degradation, which can be used for maintenance decision-making or verifying the lifetime and
reliability levels of the tested equipment.

In practice, in ADT analysis, for a given dataset, more than one model might be plausible to describe
it. Then, model uncertainty exists but in standard ADT data analysis, this is not yet fully considered
and this may lead to wrong inferences. Actually, also in accelerated life testing (ALT), different
lifetime distributions may plausibly describe the data and the problem of model uncertainty arises.
This is, for example, treated in Yu and Chang (2012) by the Bayesian model averaging (BMA)
method, with demonstration that the choice of the distribution has significant effects on the results of
the lifetime evaluation at the use conditions, especially for extreme quantiles. For ADT, the Akaike’s
information criterion (AIC) has been introduced to select the appropriate model (Park and Padgett,
2006; Park and Padgett, 2005). However, the effect of model uncertainty on the lifetime evaluation
results is not considered. In Pan and Balakrishnan (2010), both Wiener and Gamma process models
have been used, and shown to give accurate parameters estimates. However, the question remains to
how the degradation model affects the lifetime evaluation results and how the model uncertainty can
be accounted for.

As to the acceleration models, arguments are given in Yu and Chang (2012) for it not to be of
concern for the model uncertainty issue, since one can choose it based on physical considerations.

With respect to model uncertainty in literature, many works have addressed this issue (Laskey, 1996;
Nislsen and Aven, 2003; Aven and Zio, 2011; Liu et al., 2013). In the work of Zio and Apostolakis
(1996), two approaches, i.e. alternate-hypotheses (also known as model averaging) and adjustment-
factor, have been used to treat model uncertainty by expert judgments. The former one combines all
the available models through a mixture of probabilities. The latter selects a best model as reference
and updates it with information from the other models. Model averaging has been fully extended into
Bayesian model averaging (Hoeting et al., 1999; Park and Grandhi, 2014), by the integration of
model prior knowledge and the likelihood function of the obtained data for each model. In Droguett
and Mosleh (2008), it is applied to account for model uncertainty based on differences between
experimental observations and model predictions. One problem of the model averaging method is
that it assumes that the real model is one of the candidate models since the summation of model
probabilities must be equal to one, even though it is unknown and of difficult interpretation, and
controversial in practical applications. To relax this assumption, Dempster-Shafer (D-S) theory can
be introduced, using belief and plausibility functions to account for model uncertainty (Baraldi and
Zio, 2010; Park and Granhi, 2012). Some difficulties may arise in this method, in the elicitation of
expert knowledge on the belief values to assign to the models. As to the adjustment-factor method,
some work has been done for its application to accelerated testing models based on field lifetime or
degradation data. For example, Pan (2009) chose exponential and Weibull lifetime distributions as
the reference models for Device-A ALT data; then, a calibration factor is introduced to update the
reference models with the field failure data since the lab-test environment and field conditions are
different. A similar procedure is used for ADT models in (Wang et al., 2013).

4. Prognostics and Health Management

Prognostics and Health Management (PHM) is a field of research and application aiming at detecting
the degradation of engineering components, diagnosing the type of faults, predicting the failure times
and proactively managing their failures.

In the last ten years, it has drawn great attention from both the research and practice points of view,
because of the increased capacity in monitoring and the significant advancements in the techniques
of signal and data analysis, including data mining and artificial intelligence, which enable the
intelligent reading of the recorded signals and data for fault detection and diagnostics, and failure
prediction.

The objective of PHM is to take past, present and (predicted) future information on the conditions of
engineering components and systems, and use it for detecting their degradation, diagnosing their
faults, predicting their future health state evolution and their Remaining Useful Life (RUL, namely
how much time the engineering component or system can keep performing its intended function).
Ideally, the accurate prediction of the future evolution of the component health state allows running
it as long as it is healthy, with the possibility of planning maintenance work at the most convenient
and inexpensive time. Thus, expected benefits of PHM are increased plant reliability and availability,
increased equipment lifetime, increased plant safety, fewer accidents with negative impact on
environment, and optimized spare parts handling (Zio, 2012).

Traditional PHM methods have been firstly developed and verified considering single engineering
components and systems operating in fixed and stationary working conditions. For example, several
diagnostic methods have been developed and successfully applied to bearings working at fixed and
constant rotating speed and load, and prognostic methods are available for the prediction of the RUL
of structures operating under constant loads at constant temperatures (Wu et al., 2012; Zhou et al.,
2012). However, in practical industrial applications, there are a lot of engineering systems operating
in an Evolving Environment (EE), characterized by continuous or periodic variations of the working
conditions. For example, bearings in automotive vehicles are subject to continuous variation of the
rotating speed and several structures work under variable loads. The main issue of applying
traditional PHM in an EE is that the information available at the time of developing the PHM models
usually does not cover all the possible future working conditions that the engineering components
and systems will experience during their life. Thus, the traditional PHM approaches, where the
predictive models are developed using information collected in a limited set of working conditions,
tend to provide unsatisfactory performances when applied in the realistic EE conditions encountered
in practice.

The problem of developing models able to provide satisfactory performances in presence of an EE is

encountered in several research fields such as network monitoring, web mining, telecommunications
and financial data management (Dyer et al., 2014). Most engineering components and systems are
operating in an EE. For example, bearings, gears, alternators, shafts and pumps in automotive
vehicles typically work in continuously varying conditions of loads (Zhao et al., 2015; Bian et al.,
2013; Liu et al., 2014). Thus, in order to avoid a remarkable reduction of the performance of the
PHM models when they are used in an EE, it is fundamental to develop PHM models able to
continuously and automatically update themselves.

From a statistical point of view, data collected in an evolving environment are affected by a concept
drift, i.e., the statistical properties of the data change over time in an unforeseen way (Elwell and
Polikar, 2014). Since the relationship between model input and output tends to continuously change,
learning a model in presence of a concept drift is a difficult task. A possible approach to this problem
is based on the following two steps:

i. detection of the occurrence of the concept drift;

ii. once the concept drift is detected, updating the model using new input-output data collected in
the new environment.
With respect to i), the task of concept drift detection is similar to that of fault detection, i.e. verifying
whether there is a significant difference of the statistical properties between current and past signal
values. Thus, typical residual-based approaches to fault detection such as Sequential Probability
Ratio Test (SPRT) and Auto-Associative Kernel Regression (AAKR) can be applied also to the
problem of concept drift detection.

With respect to ii), some algorithms able to online update the models in order to adapt them to the
EE have been developed (Dries and Ruckert, 2009). Algorithms based on the ideas of domain
adaptation and semi-supervised learning are capable to extract information from the drifted data,
whereas active learning algorithms allow reducing the retraining cost by selecting the most
informative patterns in the drifted data (Zliobaite et al., 2011). In Razavi-Far et al. (2012), an
ensemble approach is proposed, where a new model is developed and added to the ensemble each
time a new set of data becomes available.

With respect to the problem of fault detection in an EE, the standard methods may provide
unsatisfactory performances because modifications of the working conditions can be confused with
abnormal conditions, since they may have similar effects on the behavior of the measured signals.
Thus, fault detection systems can provide several false alarms when they are used in an EE. For this
reason, the current industrial practice is to periodically update the fault detection systems in order to
limit the effects of the EE.

In the presence of an EE, empirical diagnostic models like K-Nearest Neighbors (KNN), Artificial
Neural Networks (ANN), Support Vector Machines (SVM) and Relevance Vector Machines (RVM)
may suffer a reduction of diagnostic performance. As mentioned earlier, the main reason is that the
data used to develop them are collected in a limited set of working conditions, not sufficiently
covering all the possible working conditions that may be experienced by the engineering system
during its life. Thus, diagnostic models need to learn the modification of the mapping between the
measured signals and the fault type caused by the EE. Notice that many methods of fault diagnostics
in an EE are supervised and, thus, require the availability of input-output labeled data, i.e. the
knowledge of the signal values and the corresponding fault class, whereas, in practice, the
identification of the class of the fault causing the malfunctioning in an EE is often not feasible, or
very expensive and time consuming. Thus, novel drift learning algorithms which do not rely on
labeled data are strongly needed for fault diagnostics in an EE.

With respect to the problem of prognostics in an EE, many studies have adopted approaches based
on filter algorithms, such as the Kalman and Particle Filters, which model the effect of the EE on the
equipment degradation as a process noise. In practice, the degradation measurements which
progressively become available are used to estimate the engineering system degradation state and its
future evolution, taking into account the effect of measurement and process noises. However, these
approaches are difficult to apply in situations characterized by scarce information on the degradation
process, since they require the knowledge of a physics-based degradation model and its parameters.
A further problem is that the presence of an EE can cause major modifications on the signal behavior
which cannot be described by a process noise.

A further challenge for PHM methods is their scaling up to the fleet-level needed by the modern
complex systems, comprising a large number of different components and integrating their
functionalities. Indeed, the currently proposed approaches mainly consider component- or system-
level prognostics. In such settings for PHM, it can be expected that detections, diagnostics and
predictions with different degrees of uncertainty need to be aggregated for optimal maintenance
decisions under various constraints, including resources and availability. For such distributed
maintenance management problems, some approaches are emerging, with a formulation of the
problem as multi-agent (Lei and Zuo, 2009) or holonic systems (Mavromatidi et al., 2013), but
limiting to the integration and exchange of information among the agents, without further interaction
and cooperation on the decision making for maintenance action, taking into account the different
levels of uncertainty of the different components detections, diagnostics and predictions. On the
contrary, a truly effective, distributed intelligent dynamic maintenance management system based on
PHM must be capable of accommodating exchange of information, aggregation, interaction and
cooperation at different levels of the system, coping with the different levels of uncertainty therein.
In such maintenance management system, the operating experience from different systems within
one fleet needs to be integrated in the learning process of all similar systems, looking for similarity
of behavior to actually exploit differences and therefore learn the unexpected from one system to
another. However, this is not a simple task to do as the operating and environmental conditions are
different from one system to another in the fleet, and the patterns of evolution will not, thus, be
directly transferable to any specific system. For this reason, the solution to the problem must be
dynamic, with coordination and organization of the information exchange among the different
algorithms and models tailored to the specific operating and environmental conditions of the
different systems.
From the above discussion, the challenge arises of implementing a distributed intelligent dynamic
maintenance management system that integrates aggregation and cooperation at different system
levels, and is capable of handling different degrees of uncertainty under realistic constraints and
changing operating and environmental conditions (EE). For this:
 agents at different system levels must be defined, characterized by their RUL, the
associated uncertainty, the required maintenance actions and the associated costs, and their
criticality with respect to the functionality and the safety of the entire system;
 different dynamics of the progression of the occurring degradation or fault must be
included in the model, to account for the different EE which can be gradual or
sudden/shock-like;
 the optimization problem must be formulated so as to ensure allowing that the objectives of
the agents at different system levels are achieved, while also satisfying the overall system
objectives and constraints.

From the industry point of view, for the practical implementation of any PHM-based maintenance
policy, be it a Condition-Based Maintenance (CBM) or a Preventive Maintenance (PrM) one, two
strategic problems need to be addressed:
 evaluation of the opportunity of adopting such advanced maintenance policies, founded on
specialized knowledge and modern technology, to improve performance compared to the
performance of ‘traditional’ Corrective Maintenance (CM) and PrM policies (Zio and
Compare, 2013);
 to choose which PHM method is best, for the system considered.
Indeed, any PHM approach has its advantages, limitations and drawbacks, so that choosing the best
approach for a given problem may be difficult. Criteria are needed to help the manager selecting the
best approach for RUL estimation, for minimizing unscheduled shutdowns, maximizing availability,
minimizing maintenance costs. A well-known criterion to guide design, operation and maintenance
choices is the return on investment (ROI) associated to the implementation of a given PHM
approach.
In literature, different methodologies have been proposed to evaluate the economic performance of a
PHM approach. The indicator for evaluating the economic benefit of a PHM approach proposed in
Feldman et al. (2009) is the ROI, calculated as follows:

−1 1
 In the equation, the avoided cost quantifies the economic benefit realized through using PHM,
while the investment cost includes all the costs related to PHM installation, support and maintenance.
Usually, the ROI of a PHM approach is calculated relative to unscheduled maintenance. In this case,
the avoided cost will be equivalent to the difference between the total life cycle cost of the
equipment under unscheduled maintenance and the total life cycle cost of the equipment using a
particular PHM approach (Si et al., 2011), as shown in equation (2):
− −

where is the life cycle cost in case of unscheduled maintenance, is the life cycle cost in
case of adopting PHM and is the investment cost of PHM implementation.
By using the expression of the avoided cost in equation (2), the ROI of a PHM approach in equation
(1) becomes:
− −
−1

Depending on the value of ROI, the decision maker can choose whether to implement the PHM
approach or not: if the ROI is positive, then it is recommended to implement the PHM approach; if
the ROI is null or negative, there is no benefit for implementing PHM and an unscheduled
maintenance strategy would be preferable because, different to the predictive maintenance strategy,
unscheduled maintenance does not require additional costs related to implementation.
Indeed, ROI analysis may be used for different purposes: selection of the best PHM approach to
implement, optimization of the setting and use of a particular PHM approach, and determination of
whether one should adopt PHM and predictive maintenance at all or another type of maintenance
(Feldman et al., 2009). Some authors have discussed in the literature the economic justification of
PHM (Feldman et al., 2008; Sandborn and Wilkinson, 2007; Zhang, 2013; Leao et al., 2008; Nilsson
et al., 2007; Chang et al., 2015; Fritzsche et al., 2014; Pecht, 2012). For example, the use of ROI for
PHM in malfunctioned displays of Boeing 737 has been presented in Feldman et al. (2008). PHM
results of linear replaceable units have been used to define maintenance policies and take related
decisions aimed at minimizing life cycle costs or maximizing availability in Sandborn and Wilkinson
(2007). Applications considering ROI for PHM of medical devices and commercial aircrafts have
been reported in Zhang (2013) and Leao et al. (2008). More specifically, the cost-benefit analysis of
PHM in the field of medical devices discussed in Zhang (2013) shows that the implementation of
PHM in medicine-dispensing products can save 36.2% of the service cost for the installed products
in eight cities (Zhang, 2013), and the methodology developed in Leao et al. (2008) evaluates the
economic benefits of PHM for application to commercial aircrafts considering four categories:
benefits of monitoring the system, benefits of prognostics, benefits of complete health management
and intangible benefits; for each type of benefit, a mathematical model is developed to quantify the
economic value of the benefit. In a second step and with the same logic, the costs of PHM are
categorized into three categories: development costs, aircraft costs and costs of PHM side effects,
with associated mathematical models for their quantification.
Life-cycle cost analysis has been performed in Nilsson et al. (2007) to improve maintenance
planning for a single wind turbine onshore and a wind farm offshore. The life cycle cost is calculated
for both unscheduled and scheduled maintenance as the sum of the cost of investment, the cost of
corrective and preventive maintenance, the cost for production loss and the cost of the remainder
value. The results show that for an entire farm of wind power systems, CBM is the most profitable
and a simple increase on the availability of the system (0.43%) obtained by the use of a Condition
Monitoring System (CMS) is sufficient to cover the costs related to implementing the CMS itself.
System monitoring and ROI analysis have been combined in Chang et al. (2015) for maintenance
planning of LED lighting systems. Prognostics has been introduced in Fritzsche et al. (2014) to
minimize maintenance costs in the airline industry.
However, the existing approaches and frameworks used to evaluate the ROI of a PHM
approach suffer from many gaps:
- they are for specific applications: they are not general and only fit the case study in which they
are applied (Feldman et al., 2009);
- they do not consider the impact of the performance of a PHM approach on the economic
benefit generated by it;
- they only provide a point value estimation of the ROI, without taking into account the
uncertainties in the values of the model parameters (Feldman et al., 2009).
Thus, a more general framework to calculate the ROI of a PHM approach needs to be established,
relating the ROI to different indicators of the performance of a PHM approach, which is described by
characteristics of the prediction distribution obtained by it. Such framework can be used to set PHM
requirements, i.e. the proper values of the PHM performance indicators that are required to achieve
the ROI goal. In this view, it can provide support for managers: for a fixed minimal value (threshold)
of ROI required to achieve PHM benefits, the values to be achieved by the performance indicators
for obtaining this threshold can be evaluated.

In all generality, the ROI of the PHM approach is calculated by equation (1) above. The
avoided cost is the gain from improvements of availability, reliability, maintainability and avoidance
of failures, thanks to the PHM approach implementation. It includes the costs of failures avoided and
the minimization of the loss of the RUL due to the fact that the component or system is replaced
before it is fully exploited. The investment costs are the costs associated with the realization of PHM,
the technologies and support necessary to integrate and incorporate PHM into the component or
system.
 The value of the cost avoided in equation (1) depends on the effectiveness of the PHM
approach used to predict the RUL. This depends on the degree to which the predictions are precise
and reliable. With respect to this, the ability of a model to predict accurately and with confidence the
RUL of a component or system has become more and more critical for the development of an
appropriate, timely, and cost-benefitting maintenance scheme (Kan et al., 2015). The selection of the
most appropriate PHM model for practical implementation in the field of interest relies on the
possibility of the user to evaluate the prognostic performance of the available models, used in fields
and applications similar to the one in question (Sirkoska et al., 2011). To quantify such prognostic
performance, a set of Prognostic Performance Indicators (PPIs), simple to calculate and intuitive to
understand, is needed for the comparative analysis of the different models leading to the choice of
the one to be implemented (Saxena et al., 2008).
However, as of today, prognostic models lack standard definitions of their prognostic performances
by widely agreed PPIs, because these latter suffer of inconsistent interpretations mainly due to varied
applications and domain information (Uckun et al., 2008; Sharp, 2013). A common approach to
define the PPIs of a model needs to be developed, starting from the PPIs that have been proposed in
literature for a variety of field applications.
For evaluating the capabilities of prognostic models, by PPIs quantification, it is useful to classify
the prognostic models in RUL-based and degradation-based models. Despite that for both classes of
models the final goal is the estimation of the RUL they, however, differ in the process of determining
the RUL, as in the first case it is calculated directly from the raw data, whereas in the second case it
is calculated indirectly on the basis of a Health Index (HI), constructed based on raw data (Saxena et
al., 2014). In other words, in the first case, the raw data is processed and used directly for prognostic
purposes in the RUL-based model; in the second case, the raw data is manipulated to obtain a HI,
which is a quantifiable characteristic of the equipment of interest that aggregates the relevant features
and operational conditions identifying its health (Saxena et al., 2008). The HI is, then, used in a
degradation-based model, to retrieve the RUL of the equipment.
Whilst for the RUL-based models, their prognostic performance can be directly quantified by
looking for the estimated RUL characteristics, for the degradation-based models, the prognostic
performance is dependent also on the HI characteristics. In fact, for a HI to be useful for RUL
prediction, it must have certain features and characteristics, such as being monotonic and with
defined trend. After the HI is assessed to be suitable for being used for RUL prediction, the
degradation-based models can be used and their PPIs coherently quantified.
Based on these premises, different classes of PPIs exist: PPIs for RUL performance quantification
and for HI performance quantification. Both PPIs for RUL and HI can manipulate either
instantaneous information at a given instant t or the integral information from the moment of interest
to the end of the predicted life.
Irrespective of the instantaneous or integral manipulation of the available information, the PPIs can
be primarily classified according to their characteristics: i) Accuracy, ii) Precision, iii) Stability and
iv) Spill-Over.
Accuracy PPIs quantify the closeness between the model output and the true value (Saxena et al.,
2014; Saxena et al., 2008; Walther and Moore, 2005). A very accurate model will have an estimated
RUL very close to the true one.
Precision PPIs measure the spread of the model output, by measuring the statistical variance and the
spread of the data attributable to statistical variability (Saxena et al., 2014; Walther and Moore,
2005). They quantify how confident the model is on the estimate and the degree to which a repetition
of the prognosis will yield the same results.
Stability PPIs quantify the models sensitivity by evaluating the models output with respect to factors
that directly affect the RUL estimation and prediction (i.e., the models input), and assess the models
ability to, in the long run, tend to the correct value, i.e., converge (Johnson et al., 2011; Saxena et al.,
2014). The stability of a model allows for an increased confidence in the models output.
Spill-Over PPIs measure the effects of varying the inputs on the models performance, by evaluating
the effects of reducing the number of features of a model (Saxena et al., 2008).
To compare the performances of various prognostic models, one would need to resort to a PPIs
aggregation technique, which would allow for a quantitative overall assessment of the prognostic
performances. The aggregation technique must enable the analyst to intuitively grasp the meaning of
the results provided by the PPIs and must deliver a clear suggestion for the prognostic model to be
chosen among the available ones.
For the purposes of sketching the idea of a ROI-based framework of evaluation, we consider the case
where the performance of the PHM approach is measured by the Precision Index (PI) , which
describes how close the predictions are clustered together and it is a measure of the narrowness of
the interval in which the RUL value is expected to really fall (the smaller the PI is, the more precise
the estimations given by the PHM approach are) and the reliability of the PHM approach can be
measured by the Risk Index (RI), defined as the probability that the RUL predicted by PHM is lower
than the threshold for preventive maintenance time Th. Mathematically, the RI will be equal to the
area below the RUL distribution curve, for RUL<Th. The higher the RI is, the less reliable the
implemented PHM is.
The cost of loss of the RUL of the component or system, (CL), depends on the difference
between the time of replacement of the component or system and its actual time of failure. This cost
is higher, the earlier the component or system is replaced. A threshold is given by the manager as the
minimal time needed for maintenance and CL can be calculated as the integral of the RUL prediction
distribution between the threshold and the mean value of the prediction, multiplied by the cost of the
remaining useful life per unit of time.
 The cost of corrective maintenance (CM) and the cost of failures avoided (FA) are calculated
with respect to the case when the RUL predicted by PHM exceeds Th. In this sense, the avoided cost
depends on the risk index. Given the avoided cost, the ROI of a PHM approach can be calculated and
a relationship can be obtained between the ROI and the characteristics of the predicted RUL
distribution.
A framework such as the one sketched above can be used to guide the selection of the adequate PHM
approach to implement in order to reach a desired ROI. In fact, the decision maker can define the
requirements for the performance metrics of the PHM approach leading to a minimal desired ROI. It
is expected that to reach a high value of ROI, the predictions should be precise and the component or
system should be repaired/replaced at a time such to avoid the failure of the equipment on one hand
but, also, minimize, on the other hand, the costs related to the loss of still exploitable RUL. Note that
since the predictions of PHM are subject to uncertainties, the framework of ROI evaluation should be
able to include the changing operational and environmental factors that may affect the RUL
predictions. Moreover, in practice also, the threshold for preventive maintenance time Th is typically
uncertain and, thus, sensitivity analysis should be performed to study how the variation of Th affects
the variation of the ROI of a given PHM approach.
The ROI–based evaluation framework sketched above is only illustrative and, in practice, to ensure
that the adopted PHM method meets the goal of confident maintenance decision making, the
prognostic method upon which to rely for RUL prediction should not only satisfy the requirements
on the expected quality of the RUL point estimate (e.g., the accuracy PPIs described), on uncertainty
quantification (e.g., the precision PPIs described above), and on the other desirable characteristics
(measured by the other PPIs mentioned above), but also meet some requirements of trustworthiness
of the prognostic method itself, i.e., the credibility that for the specific RUL prediction problem of
interest, the method can be trusted to provide an accurate and precise RUL with correct and fair
uncertainty quantification.

In this view, one must assess the prediction capability of a prognostic method, i.e. the property of
providing trustable RUL predictions, with the quality characteristics required for the purpose of the
maintenance decision making of interest. The RUL quality characteristics are measured by the PPIs
defined above. For the trustworthiness of the prognostic method, one must introduce some concept of
prediction capability maturity to measure the property that a prognostic method can provide
trustworthy results, based on the inherent characteristics of the method itself in relation to RUL
predictions, and on proven experiences of application in other problem settings similar to the one
under analysis: the more mature a prognostic method, the more confidence that its predictions can be
trusted.

To evaluate the prediction capability of a prognostic method, both the RUL prediction quality (as
measured by the PPIs) and the trustworthiness of the prognostic method itself should be considered,
as illustrated in Figure 1Figure 1. As seen above, a number of approaches can be used for measuring Mis e
Roma
the quality performance of a prognostic method by evaluating its RUL predictions with respect to Mis e
whether the method can yield accurate and precise RUL estimates, considering both point estimate
and uncertainty (i.e., in terms of the PPIs). On the other hand, only few methods have been proposed
for the assessment of the trustworthiness of a prediction method or model (Paulk et al., 1993;
Oberkampf et al., 2007), and none exists in the specific context and for the specific aim of
prognostics and maintenance decision making. To fill this gap, a structured framework should be
proposed for the assessment of the prediction capability maturity of a prognostics method, to be
integrated with the RUL prediction quality, measured conditioned on the prognostic model
performance given the available data, to yield the overall prognostic method prediction capability
(Figure 1Figure 1). Mis e
Roma
 Prognostic method
prediction capability

RUL prediction
quality

Method
trustworthiness

RUL point Uncertainty

estimate quantification

Reliability Validity

Model
Mathematical
structure V.S.
modeling
data
adequacy
adequacy

AND gate
(intersection)

INHIBIT gate
(conditional)

Figure 1 Prediction capability consists of RUL prediction quality and prognostic method trustworthiness.
The former can be quantified resorting to the Prognostic Performance Indicators (PPIs), whereas the latter
should be quantified resorting to a structured prediction capability maturity assessment framework.

Finally, PHM utilizing databases and information technology is the enabling technique for
condition-based and predictive maintenance. However, in practice a challenge for the use of
PHM, and the associated condition-based and predictive maintenance policies, comes to the
reliability qualification of a component or system at the design stage, for its licensing. The
question is how to include these aspects in the reliability qualification or safety case: how
must these be changed when scheduled maintenance is replaced with condition-based and/or
predictive maintenance? This question has become fundamental, as industry wishes to move
to these more confident, effective, flexible, smart maintenance policies of their smart
components and systems, but safety and reliability demonstration are still needed and
required. The PHM techniques using condition monitoring data can contribute to the analysis
for evaluating failure frequency and probability with reduced conservatism, thus allowing for
an online reliability assessment, with reduced uncertainty. In the online, living assessment, if
the statistics-based distributions are replaced with condition-based distributions obtained by
PHM reliability estimates and uncertainty bands are no longer fixed a priori but change
depending on the predictive power of the PHM techniques employed. The frequency of
failures, the equipment unavailability, the probability of human error and the probability of
recovery a failure change due to aging effects and other environmental and operational
factors, the components and systems reliability estimates change too, and so do the associated
uncertainty bands. This requires that the estimation be done time-dependently on the basis of
the prognostics results, based on the prior assessment and a Bayesian update using the
prognostics data (Kim et al., 2015).

5. Conclusions

The reliability engineering field must continuously evolve to be at speed with the industrial
and societal developments. This requires continuous advancement of technical knowledge and
ability. In the current scenario of technological development, with strong digitalization and
interconnection at all levels of cyber-physical systems and in all industrial sectors, reliability
engineering is faced with new challenges but it is also exposed to new opportunities of
improvement.

In this paper, I have focused on the reliability modeling of components and systems, and
discussed some challenges and opportunities related to degradation modeling and prognostics
and health management. These areas of research and application hold great promise for the
improvement of the safety and the productive and service capacity of industrial components
and systems. The knowledge, information and data that are increasingly acquired offer new
ways of modeling and analysis, with significant potential benefits.

The practical implementation of these new ways needs to be followed up with care, for
effectively collecting such benefits. For example, it is clear that PHM utilizing databases and
information technology is the enabling technique for condition-based and predictive
maintenance. However, in practice a challenge for the use of PHM, and the associated
condition-based and predictive maintenance policies, comes to the reliability qualification of a
component or system that must be done a priori, at the design stage, for its licensing. The
question is how to include these aspects in the reliability qualification or safety case: how
must these be changed when scheduled maintenance is replaced with condition-based and/or
predictive maintenance? This question has become fundamental, as industry wishes to move
to these more confident, effective, flexible, smart maintenance policies of their smart
components and systems, but safety and reliability demonstration are still needed and
required.
Furthermore, the PHM techniques using condition monitoring data can contribute to the
analysis for evaluating failure frequency and probability with reduced conservatism, thus
allowing for an online reliability assessment, with reduced uncertainty. In the online, living
assessment, if the statistics-based distributions are replaced with condition-based distributions
obtained by PHM, reliability estimates and uncertainty bands are no longer fixed a priori but
change depending on the predictive power of the PHM techniques employed. As the
frequency of failures, the equipment unavailability, the probability of human error and the
probability of recovering a failure change due to aging effects and other environmental and
operational factors, the components and systems reliability estimates change too, and so do
the associated uncertainty bands. This requires that the estimation be done time-dependently
on the basis of the prognostics results, based on a prior assessment and a Bayesian update
using the prognostics data (Kim et al., 2015).

Acknowledgments

I would like to acknowledge the insightful work and sharing experience of the many colleagues which
have shared with me passionate study and research experiences on the subjects of this paper. These are
the colleagues from the Chaire on Systems Science and the Energetic Challenge (SSEC) of
CentraleSupelec, Paris, www.ssde.fr , the Laboratory of Signal Analysis and Risk Analysis (LASAR)
at the Department of Energy of the Politecnico di Milano, www.lasar.polimi.it, the Center for
Resilience and Safety of Critical Infrastructures (CRESCI), http://cresci.buaa.edu.cn/, and Prof. Olga
Fink. Deep thanks go also to Dr. Jie Liu of CentraleSupelec, Paris, for his contribution in the
preparation of the manuscript.

1. References

Agrafiotis, G. K., and M. Z. Tsoukalas. "On excess-time correlated cumulative processes." Journal of
the Operational Research Society (1995): 1269-1280.
Alam, Mansoor, and Ubaid M. Al-Saggaf. "Quantitative reliability evaluation of repairable phased-
mission systems using Markov approach." IEEE Transactions on Reliability 5.35 (1986): 498-
503.
Albin, S. L., and Seng Chao. "Preventive replacement in systems with dependent components."
Reliability, IEEE Transactions on 41.2 (1992): 230-238.
Anderson, Kevin K. "Limit theorems for general shock models with infinite mean intershock times."
Journal of applied probability (1987): 449-456.
Aven, Terje, and Enrico Zio. "Some considerations on the treatment of uncertainties in risk
assessment for practical decision making." Reliability Engineering & System Safety 96.1
(2011): 64-74.
Bai, Jian-Ming, Ze-Hui Li, and Xin-Bing Kong. "Generalized shock models based on a cluster point
process." Reliability, IEEE Transactions on 55.3 (2006): 542-550.
Baraldi, Piero, and Enrico Zio. "A Comparison Between Probabilistic and Dempster‐Shafer Theory
Approaches to Model Uncertainty Analysis in the Performance Assessment of Radioactive
Waste Repositories." Risk Analysis 30.7 (2010): 1139-1156.
Barros, Anne, Christophe Berenguer, and Antoine Grall. "A maintenance policy for two-unit parallel
systems based on imperfect monitoring information." Reliability Engineering & System Safety
91.2 (2006): 131-136.
Becker, Günter, Leonidas Camarinopoulos, and Dimitris Kabranis. "Dynamic reliability under random
shocks." Reliability Engineering & System Safety 77.3 (2002): 239-251.
Bian, Linkan, and Nagi Gebraeel. "Stochastic methodology for prognostics under continuously varying
environmental profiles." Statistical Analysis and Data Mining 6.3 (2013): 260-270.
Black, M., A. T. Brint, and J. R. Brailsford. "A semi-Markov approach for modelling asset deterioration."
Journal of the Operational Research Society 56.11 (2005): 1241-1249.
Bondavalli, Andrea, et al. "Dependability modeling and evaluation of multiple-phased systems using
DEEM." Reliability, IEEE Transactions on 53.4 (2004): 509-522.
Castanier, Bruno, Antoine Grall, and Christophe Bérenguer. "A condition-based maintenance policy
with non-periodic inspections for a two-unit series system." Reliability Engineering & System
Safety 87.1 (2005): 109-120.
Chalk, Steven G., and James F. Miller. "Key challenges and recent progress in batteries, fuel cells,
and hydrogen storage for clean energy systems." Journal of Power Sources 159.1 (2006): 73-
80.
Chang, Moon-Hwan, et al. "A return on investment analysis of applying health monitoring to LED
lighting systems." Microelectronics Reliability 55.3 (2015): 527-537.
Chen, Nan, et al. "Condition-based maintenance using the inverse Gaussian degradation model."
European Journal of Operational Research 243.1 (2015): 190-199.
Chookah, Mohamed, Mohammad Nuhi, and Mohammad Modarres. "A probabilistic physics-of-failure
model for prognostic health management of structures subject to pitting and corrosion-fatigue."
Reliability Engineering & System Safety 96.12 (2011): 1601-1610.
Daigle, Matthew J., and Kai Goebel. "A model-based prognostics approach applied to pneumatic
valves." International Journal of Prognostics and Health Management 2 (2011): 84.
Liu, Dong, et al. "Reliability analysis of phased-mission systems using Bayesian networks." Reliability
and Maintainability Symposium, 2008. RAMS 2008. Annual. IEEE, 2008.
Dekker, Rommert, Raymond Ph Plasmeijer, and Jan H. Swart. "Evaluation of a new maintenance
concept for the preservation of highways." IMA Journal of Management Mathematics 9.2
(1998): 109-156.
Dries, Anton, and Ulrich Rückert. "Adaptive concept drift detection." Statistical Analysis and Data
Mining 2.5‐6 (2009): 311-327.
Droguett, Enrique López, and Ali Mosleh. "Bayesian methodology for model uncertainty using model
performance data." Risk Analysis 28.5 (2008): 1457-1476.
Dugan, JoAnne Bechta. "Automated analysis of phased-mission reliability." Reliability, IEEE
Transactions on 40.1 (1991): 45-52.
Dyer, Karl B., Robert Capo, and Robi Polikar. "Compose: A semisupervised learning framework for
initially labeled nonstationary streaming data." Neural Networks and Learning Systems, IEEE
Transactions on 25.1 (2014): 12-26.
Elwell, Ryan, and Robi Polikar. "Incremental learning of concept drift in nonstationary environments."
Neural Networks, IEEE Transactions on 22.10 (2011): 1517-1531.
Esary, J. D., and A. W. Marshall. "Shock models and wear processes." The annals of probability
(1973): 627-649.
Escobar, Luis A., and William Q. Meeker. "A review of accelerated test models." Statistical Science
(2006): 552-577.
Fan, Juanjuan, S. G. Ghurye, and Richard A. Levine. "Multicomponent lifetime distributions in the
presence of ageing." Journal of applied probability 37.2 (2000): 521-533.
Feldman, Kiri, Peter Sandborn, and Taoufik Jazouli. "The analysis of return on investment for PHM
applied to electronic systems." Prognostics and Health Management, 2008. PHM 2008.
International Conference on. IEEE, 2008.
Feldman, Kiri, Taoufik Jazouli, and Peter A. Sandborn. "A methodology for determining the return on
investment associated with prognostics and health management." Reliability, IEEE
Transactions on 58.2 (2009): 305-316.
Fritzsche, R., J. N. D. Gupta, and R. Lasch. "Optimal prognostic distance to minimize total
maintenance cost: The case of the airline industry." International Journal of Production
Economics 151 (2014): 76-88.
Fussel J. B., et al. “Phased-Mission System Reliability Analysis,” Volume 1: Methodology. Electric
Power Research Institute Research Report 1233-2. University of Tennessee, Knoxville
Tennessee, July 1981.
Gebraeel, Nagi, Alaa Elwany, and Jing Pan. "Residual life predictions in the absence of prior
degradation knowledge." Reliability, IEEE Transactions on 58.1 (2009): 106-117.
Giorgio, Massimiliano, Maurizio Guida, and Gianpaolo Pulcini. "An age-and state-dependent Markov
model for degradation processes." IIE Transactions 43.9 (2011): 621-632.
Hoeting, Jennifer A., et al. "Bayesian model averaging: a tutorial." Statistical science (1999): 382-401.
Hong, H. P., et al. "Optimal condition-based maintenance decisions for systems with dependent
stochastic degradation of components." Reliability Engineering & System Safety 121 (2014):
276-288.
Kim, Hyeonmin, et al. "Reliability data update using condition monitoring and prognostics in
probabilistic safety assessment." Nuclear Engineering and Technology 47.2 (2015): 204-211.
Jiang, Lei, Qianmei Feng, and David W. Coit. "Reliability and maintenance modeling for dependent
competing failure processes with shifting failure thresholds." Reliability, IEEE Transactions on
61.4 (2012): 932-948.
Johnson, Stephen B., et al., eds. System health management: with aerospace applications. John
Wiley & Sons, 2011.
Kan, Man Shan, Andy CC Tan, and Joseph Mathew. "A review on prognostic techniques for non-
stationary and non-linear rotating systems." Mechanical Systems and Signal Processing 62
(2015): 1-20.
Keedy, Elias, and Qianmei Feng. "A physics-of-failure based reliability and maintenance modeling
framework for stent deployment and operation." Reliability Engineering & System Safety 103
(2012): 94-101.
Klutke, Georgia-Ann, and Yoonjung Yang. "The availability of inspected systems subject to shocks
and graceful degradation." Reliability, IEEE Transactions on 51.3 (2002): 371-374.
Kim, Kuk, and Kyung S. Park. "Phased-mission system reliability under Markov environment."
Reliability, IEEE Transactions on 43.2 (1994): 301-309.
Kröger, Wolfgang, and Enrico Zio. Vulnerable systems. Springer Science & Business Media, 2011.
Lai, Min-Tsai, and Ying-Chang Chen. "Optimal periodic replacement policy for a two-unit system with
failure rate interaction." The international journal of advanced manufacturing technology 29.3-
4 (2006): 367-371.
Laskey, Kathryn Blackmond. "Model uncertainty: Theory and practical implications." Systems, Man
and Cybernetics, Part A: Systems and Humans, IEEE Transactions on 26.3 (1996): 340-348.
Lawless, Jerry, and Martin Crowder. "Covariates and random effects in a gamma process model with
application to degradation and failure." Lifetime Data Analysis 10.3 (2004): 213-227.
Leão, Bruno P., et al. "Cost-benefit analysis methodology for PHM applied to legacy commercial
aircraft." Aerospace Conference, 2008 IEEE. IEEE, 2008.
Lei, Yaguo, and Ming J. Zuo. "Gear crack level identification based on weighted K nearest neighbor
classification algorithm." Mechanical Systems and Signal Processing 23.5 (2009): 1535-1547.
Li, Wenjian, and Hoang Pham. "Reliability modeling of multi-state degraded systems with multi-
competing failures and random shocks." Reliability, IEEE Transactions on 54.2 (2005): 297-
303.
Li, Yan-Fu, Enrico Zio, and Yan-Hui Lin. "A multistate physics model of component degradation based
on stochastic petri nets and simulation." Reliability, IEEE Transactions on 61.4 (2012): 921-
931.
Liao, Haitao, and Elsayed A. Elsayed. "Reliability inference for field conditions from accelerated
degradation testing." Naval Research Logistics (NRL) 53.6 (2006): 576-587.
Levitin, G., and A. Lisnianski. "Multi-state system reliability: assessment, optimization and
applications." (2003).
Lim, Heonsang, and Bong-Jin Yum. "Optimal design of accelerated degradation tests based on
Wiener process models." Journal of Applied Statistics 38.2 (2011): 309-325.
Lin, Yan-Hui, Yan-Fu Li, and Enrico Zio. "Fuzzy reliability assessment of systems with multiple-
dependent competing degradation processes." Fuzzy Systems, IEEE Transactions on 23.5
(2015): 1428-1438.
Ling, Man Ho, Kwok Leung Tsui, and Narayanaswamy Balakrishnan. "Accelerated degradation
analysis for the quality of a system based on the Gamma process." Reliability, IEEE
Transactions on 64.1 (2015): 463-472.
Liu, Jie, et al. "Nuclear power plant components condition monitoring by probabilistic support vector
machine." Annals of Nuclear Energy 56 (2013): 23-33.
Liu, Jie, et al. "AN EFFICIENT ONLINE LEARNING APPROACH FOR SUPPORT VECTOR
REGRESSION." Decision Making and Soft Computing: Proceedings of the 11th International
FLINS Conference. Vol. 10. World Scientific, 2014.
Lu, C. Joseph, and William O. Meeker. "Using degradation measures to estimate a time-to-failure
distribution." Technometrics 35.2 (1993): 161-174.
Lu, Jye-Chyi, Jinho Park, and Qing Yang. "Statistical inference of a time-to-failure distribution derived
from linear degradation data." Technometrics 39.4 (1997): 391-400.
Lydell, Bengt OY. "Pipe failure probability—the Thomas paper revisited." Reliability Engineering &
System Safety 68.3 (2000): 207-217.
Mavromatidis, Georgios, Salvador Acha, and Nilay Shah. "Diagnostic tools of energy performance for
supermarkets using Artificial Neural Network algorithms." Energy and Buildings 62 (2013):
304-314.
Mura, Ivan, and Andrea Bondavalli. "Hierarchical modeling and evaluation of phased-mission
systems." Reliability, IEEE Transactions on 48.4 (1999): 360-368.
Meeker, William Q., and Luis A. Escobar. Statistical methods for reliability data. John Wiley & Sons,
2014.
Moghaddass, Ramin, and Ming J. Zuo. "Multistate degradation and supervised estimation methods for
a condition-monitored device." IIE Transactions 46.2 (2014): 131-148.
Murthy, D. N. P., and D. G. Nguyen. "Study of a multi-component system with failure interaction."
European Journal of Operational Research 21.3 (1985): 330-338.
Mura, Ivan, and Andrea Bondavalli. "Markov regenerative stochastic Petri nets to model and evaluate
phased mission systems dependability." IEEE Transactions on Computers 12 (2001): 1337-
1351.
Musallam, Mahera, et al. "Application of coupled electro-thermal and physics-of-failure-based analysis
to the design of accelerated life tests for power modules." Microelectronics Reliability 54.1
(2014): 172-181.
Pecht, Michael. "Nvidia’s GPU failures: A case for prognostics and health management."
Microelectronics Reliability 52.6 (2012): 953-957.
Nelson, Wayne B. Accelerated testing: statistical models, test plans, and data analysis. Vol. 344. John
Wiley & Sons, 2009.
Nicolai, Robin P., and Rommert Dekker. Optimal maintenance of multi-component systems: a review.
Springer London, 2008.
Nilsen, Thomas, and Terje Aven. "Models and model uncertainty in the context of risk analysis."
Reliability Engineering & System Safety 79.3 (2003): 309-317.
Nilsson, Julia, and Lina Bertling. "Maintenance management of wind power systems using condition
monitoring systems—life cycle cost analysis for two case studies." Energy Conversion, IEEE
Transactions on 22.1 (2007): 223-229..
Özekici, Süleyman. "Optimal periodic replacement of multicomponent reliability systems." Operations
Research 36.4 (1988): 542-552.
Oberkampf, William L., Martin Pilch, and Timothy G. Trucano. Predictive capability maturity model for
computational modeling and simulation. No. SAND2007-5948. Sandia National Laboratories,
2007.
Pan, Rong. "A Bayes approach to reliability prediction utilizing data from accelerated life tests and field
failure observations." Quality and Reliability Engineering International 25.2 (2009): 229-240.
Pan, Zhengqiang, and N. Balakrishnan*. "Multiple-steps step-stress accelerated degradation modeling
based on Wiener and gamma processes." Communications in Statistics-Simulation and
Computation 39.7 (2010): 1384-1402.
Park, Chanseok, and William J. Padgett. "Stochastic degradation models with several accelerating
variables." Reliability, IEEE Transactions on 55.2 (2006): 379-390.
Park, Inseok, and Ramana V. Grandhi. "Quantification of model-form and parametric uncertainty using
evidence theory." Structural Safety 39 (2012): 44-51.
Park, Inseok, and Ramana V. Grandhi. "A Bayesian statistical method for quantifying model form
uncertainty and two model combination methods." Reliability Engineering & System Safety
129 (2014): 46-56.
Park, Chanseok, and W. J. Padgett. "Accelerated degradation models for failure based on geometric
Brownian motion and gamma processes." Lifetime Data Analysis 11.4 (2005): 511-527.
Paulk, Mark C., et al. "Capability maturity model, version 1.1." Software, IEEE 10.4 (1993): 18-27.
Peng, Weiwen, et al. "Inverse Gaussian process models for degradation analysis: A Bayesian
perspective." Reliability Engineering & System Safety 130 (2014): 175-189.
Peng, Hao, Qianmei Feng, and David W. Coit. "Reliability and maintenance modeling for systems
subject to multiple dependent competing failure processes." IIE transactions 43.1 (2010): 12-
22.
Rafiee, Koosha, Qianmei Feng, and David W. Coit. "Reliability modeling for dependent competing
failure processes with changing degradation rate." IIE transactions 46.5 (2014): 483-496.
Rasmekomen, Nipat, and Ajith Kumar Parlikad. "Maintenance optimization for asset systems with
dependent performance degradation." Reliability, IEEE Transactions on 62.2 (2013): 362-367.
Razavi-Far, Roozbeh, Piero Baraldi, and Enrico Zio. "Dynamic weighting ensembles for incremental
learning and diagnosing new concept class faults in nuclear power systems." Nuclear Science,
IEEE Transactions on 59.5 (2012): 2520-2530.
Reggiani, Susanna, et al. "Physics-based analytical model for HCS degradation in STI-LDMOS
transistors." Electron Devices, IEEE Transactions on 58.9 (2011): 3072-3080.
Salonen, Jorma, et al. "Experience on in-service damage in power plant components." Engineering
Failure Analysis 14.6 (2007): 970-977.
Sandborn, Peter A., and Chris Wilkinson. "A maintenance planning and business case development
model for the application of prognostics and health management (PHM) to electronic
systems." Microelectronics Reliability 47.12 (2007): 1889-1901.
Santini, Thomas, et al. "Accelerated degradation data of SiC MOSFETs for lifetime and Remaining
 Useful Life assessment." Microelectronics Reliability 54.9 (2014): 1718-1723.
Saxena, Abhinav, Shankar Sankararaman, and Kai Goebel. "Performance evaluation for fleet-based
and unit-based prognostic methods." Second European conference of the Prognostics and
Health Management society. 2014.
Saxena, Abhinav, et al. "Metrics for evaluating performance of prognostic techniques." Prognostics
and health management, 2008. phm 2008. international conference on. IEEE, 2008.
Sharp, Michael E. "Simple Metrics for Evaluating and Conveying Prognostic Model Performance To
Users With Varied Backgrounds." In other words 1 (2013): 1.
Si, Xiao-Sheng, et al. "Remaining useful life estimation–a review on the statistical data driven
approaches." European Journal of Operational Research 213.1 (2011): 1-14.
Sikorska, J. Z., Melinda Hodkiewicz, and Lin Ma. "Prognostic modelling options for remaining useful
life estimation by industry." Mechanical Systems and Signal Processing 25.5 (2011): 1803-
1836.
Smotherman, Mark, and Kay Zemoudeh. "A non-homogeneous Markov model for phased-mission
reliability analysis." Reliability, IEEE Transactions on 38.5 (1989): 585-590.
Song, Sanling, et al. "Reliability analysis for multi-component systems subject to multiple dependent
competing failure processes." Reliability, IEEE Transactions on 63.1 (2014): 331-345.
Song, Sanling, David W. Coit, and Qianmei Feng. "Reliability for systems of degrading components
with distinct component shock sets." Reliability Engineering & System Safety 132 (2014): 115-
124.
Somani, Arun K., James A. Ritcey, and Stephen HL Au. "Computationally-efficient phased-mission
reliability analysis for systems with variable configurations." Reliability, IEEE Transactions on
41.4 (1992): 504-511.
Straub, Daniel. "Stochastic modeling of deterioration processes through dynamic Bayesian networks."
Journal of Engineering Mechanics 135.10 (2009): 1089-1099.
Thomas, L. C. "A survey of maintenance and replacement models for maintainability and reliability of
multi-item systems." Reliability Engineering 16.4 (1986): 297-309.
Thomas, E. V., et al. "Statistical methodology for predicting the life of lithium-ion cells via accelerated
degradation testing." Journal of Power Sources 184.1 (2008): 312-317.
Tseng, Sheng-Tsaing, Narayanaswamy Balakrishnan, and Chih-Chun Tsai. "Optimal step-stress
accelerated degradation test plan for gamma degradation processes." Reliability, IEEE
Transactions on 58.4 (2009): 611-618.
Uckun, Serdar, Kai Goebel, and Peter JF Lucas. "Standardizing research methods for prognostics."
Prognostics and Health Management, 2008. PHM 2008. International Conference on. IEEE,
2008.
Unwin, Stephen D., et al. Multi-state physics models of aging passive components in probabilistic risk
assessment. No. PNNL-SA-76893. Pacific Northwest National Laboratory (PNNL), Richland,
WA (US), 2011.
van Dijkhuizen, Gerhard. "Maintenance grouping in multi-step multi-component production systems."
Maintenance, modeling and optimization. Springer US, 2000. 283-306.
Van Horenbeek, Adriaan, and Liliane Pintelon. "A dynamic predictive maintenance policy for complex
multi-component systems." Reliability Engineering & System Safety 120 (2013): 39-50.
Walther, Bruno A., and Joslin L. Moore. "The concepts of bias, precision and accuracy, and their use
in testing the performance of species richness estimators, with a literature review of estimator
performance." Ecography 28.6 (2005): 815-829.
Wang, Chaonan, Liudong Xing, and Gregory Levitin. "Competing failure analysis in phased-mission
systems with functional dependence in one of phases." Reliability Engineering & System
Safety 108 (2012): 90-99.
Wang, Yaping, and Hoang Pham. "Modeling the dependent competing risks with multiple degradation
processes and random shock using time-varying copulas." Reliability, IEEE Transactions on
 61.1 (2012): 13-22.
Wang, Xiao, and Dihua Xu. "An inverse Gaussian process model for degradation data."
Technometrics 52.2 (2010): 188-197.
Wang, Chaonan, Liudong Xing, and Gregory Levitin. "Reliability analysis of multi-trigger binary
systems subject to competing failures." Reliability Engineering & System Safety 111 (2013): 9-
17.
Wang, Fu-Kwun, and Tao-Peng Chu. "Lifetime predictions of LED-based light bars by accelerated
degradation test." Microelectronics Reliability 52.7 (2012): 1332-1336.
Wang, Lizhi, et al. "A Bayesian reliability evaluation method with integrated accelerated degradation
testing and field information." Reliability Engineering & System Safety 112 (2013): 38-47.
Wang, Zhonglai, et al. "An approach to reliability assessment under degradation and shock process."
Reliability, IEEE Transactions on 60.4 (2011): 852-863.
Whitmore, G. A. "Estimating degradation by a Wiener diffusion process subject to measurement error."
Lifetime data analysis 1.3 (1995): 307-319.
Whitmore, G. A., and Fred Schenkelberg. "Modelling accelerated degradation data using Wiener
diffusion with a time scale transformation." Lifetime data analysis 3.1 (1997): 27-45.
Wortman, M. A., Georgia-Ann Klutke, and Hayriye Ayhan. "A maintenance strategy for systems
subjected to deterioration governed by random shocks." Reliability, IEEE Transactions on 43.3
(1994): 439-445.
Wu, Shuen-De, et al. "Bearing fault diagnosis based on multiscale permutation entropy and support
vector machine." Entropy 14.8 (2012): 1343-1356.
Xing, Liudong, and Joanne Bechta Dugan. "Analysis of generalized phased-mission system reliability,
performance, and sensitivity." Reliability, IEEE Transactions on 51.2 (2002): 199-211.
Xing, Liudong, and Joanne Bechta Dugan. "A separable ternary decision diagram based analysis of
generalized phased-mission reliability." Reliability, IEEE Transactions on 53.2 (2004): 174-
184.
Xing, Liudong, and Gregory Levitin. "Combinatorial analysis of systems with competing failures subject
to failure isolation and propagation effects." Reliability Engineering & System Safety 95.11
(2010): 1210-1215.
Xing, Liudong, and Gregory Levitin. "BDD-based reliability evaluation of phased-mission systems with
internal/external common-cause failures." Reliability Engineering & System Safety 112 (2013):
145-153.
Xing, Liudong, Chaonan Wang, and Gregory Levitin. "Competing failure analysis in non-repairable
binary systems subject to functional dependence." Proceedings of the Institution of
Mechanical Engineers, Part O: Journal of Risk and Reliability 226.4 (2012): 406-416.
Yang, Guangbin. "Environmental-stress-screening using degradation measurements." Reliability, IEEE
Transactions on 51.3 (2002): 288-293.
Yang, Won Seok, Dae E. Lim, and Kyung C. Chae. "Maintenance of multi-state production systems
deteriorated by random shocks and production." Journal of Systems Science and Systems
Engineering 20.1 (2011): 110-118.
Yang, Qingyu, Nailong Zhang, and Yili Hong. "Reliability analysis of repairable systems with
dependent component failures under partially perfect repair." Reliability, IEEE Transactions on
62.2 (2013): 490-498.
Yang, Kai, and Guangbin Yang. "Degradation reliability assessment using severe critical values."
International Journal of Reliability, Quality and Safety Engineering 5.01 (1998): 85-95.
Ye, Zhi Sheng, Loon Ching Tang, and Hai Yan Xu. "A distribution-based systems reliability model
under extreme shocks and natural degradation." Reliability, IEEE Transactions on 60.1 (2011):
246-256.
Ye, Zhi‐Sheng, and Min Xie. "Stochastic modelling and analysis of degradation for highly reliable
products." Applied Stochastic Models in Business and Industry 31.1 (2015): 16-32.
Ye, Zhi-Sheng, and Nan Chen. "The inverse Gaussian process as a degradation model."
Technometrics 56.3 (2014): 302-311.
Ye, Zhi-Sheng, et al. "Accelerated degradation test planning using the inverse Gaussian process."
Reliability, IEEE Transactions on 63.3 (2014): 750-763.
Ou, Yong, and Joanne Bechta Dugan. "Modular solution of dynamic multi-phase systems." Reliability,
IEEE Transactions on 53.4 (2004): 499-508.
Yu, I., and Che-Lun Chang. "Applying bayesian model averaging for quantile estimation in accelerated
life tests." Reliability, IEEE Transactions on 61.1 (2012): 74-83.
Zang, Xinyu, Hairong Sun, and Kishor S. Trivedi. "A BDD-based algorithm for reliability analysis of
phased-mission systems." Reliability, IEEE Transactions on 48.1 (1999): 50-60.
Zequeira, Rómulo I., and Christophe Bérenguer. "On the inspection policy of a two-component parallel
system with failure interaction." Reliability Engineering & System Safety 88.1 (2005): 99-107.
Zhang, Qian. "Case study of cost benefits of condition based maintenance used in medical devices."
Reliability and Maintainability Symposium (RAMS), 2013 Proceedings-Annual. IEEE, 2013.
Zhao, Fuqiong, et al. "An Integrated Prognostics Method Under Time-Varying Operating Conditions."
Reliability, IEEE Transactions on 64.2 (2015): 673-686.
Zhou, Y., et al. "Application of the horizontal slice of cyclic bispectrum in rolling element bearings
diagnosis." Mechanical Systems and Signal Processing 26 (2012): 229-243.
Žliobaitė, Indrė, et al. "Active learning with evolving streaming data." Machine Learning and
Knowledge Discovery in Databases. Springer Berlin Heidelberg, 2011. 597-612.
Zio, Enrico, and G. E. Apostolakis. "Two methods for the structured assessment of model uncertainty
by experts in performance assessments of radioactive waste repositories." Reliability
Engineering & System Safety 54.2 (1996): 225-241.
Zio, Enrico. "Prognostics and health management of industrial equipment." Diagnostics and
Prognostics of Engineering Systems: Methods and Techniques (2012): 333-356.
Zio, Enrico, and Michele Compare. "Evaluating maintenance policies by quantitative modeling and
analysis." Reliability Engineering & System Safety 109 (2013): 53-65.