Guru Prakash

Toward a Big Data-Based

Assistant Professor
Department of Civil Engineering,
Indian Institute of Technology Indore,
Approach: A Review
Indore, Madhya Pradesh 453552, India
e-mail: [email protected]
on Degradation Models
Xian-Xun Yuan
Associate Professor
for Prognosis of Critical
Department of Civil Engineering,
Ryerson University,
Infrastructure
350 Victoria Street,
Toronto, ON, M5B 2K3, Canada Safety and reliability of large critical infrastructure such as long-span bridges, high-rise
e-mail: [email protected] buildings, nuclear power plants, high-voltage transmission towers, rotating machinery,
and so on, are important for a modern society. Research on reliability and safety analysis
Budhaditya Hazra started with a “small data” problem dealing with relative scarce lifetime or failure data.
Associate Professor Later, degradation modeling that uses performance deterioration, or, condition data col-
Department of Civil Engineering, lected from in-service inspections or online health monitoring became an important tool
Indian Institute of Technology Guwahati, for reliability prediction and maintenance planning of highly reliable engineering
Guwahati 781039, India systems. Over the past decades, a large number of degradation models have been developed
e-mail: [email protected] to characterize and quantify the underlying degradation mechanism using direct and indi-
rect measurements. Recent advancements in artificial intelligence, remote sensing, big data
Daijiro Mizutani analytics, and Internet of things are making far-reaching impacts on almost every aspect of
Assistant Professor our lives. The effect of these changes on the degradation modeling, prognosis, and safety
Department of Civil and Environmental management is interesting questions to explore. This paper presents a comprehensive,
Engineering, forward-looking review of the various degradation models and their practical applications
Tohoku University, to damage prognosis and management of critical infrastructure. The degradation models
6-6-06, Aramaki, Aoba, are classified into four categories: physics-based, knowledge-based, data-driven, and
Sendai, Miyagi 980-8579, Japan hybrid approaches. [DOI: 10.1115/1.4048787]
e-mail: [email protected]
Keywords: continuous and periodic condition assessment, diagnostic feature extraction,
online diagnostic approaches

1 Introduction is in contrast to the other failure type, called hard, or, shock failures,
mainly caused by shocks or catastrophic events such as earthquake,
Civil infrastructure and mechanical systems such as buildings,
tsunami, hurricane, tornado, flash flood, explosion, and terrorism.
bridges, power plants, and machinery are an integral part of
While the prediction of hard failures is an equally, if not more,
modern society. Many of these systems operate under harsh operat-
important and challenging task, this paper focuses on the soft fail-
ing and loading conditions, a significant percentage nearing their
ures defined by one or multiple degradation mechanisms with pre-
end of design life. This combination of factors often leads to a
defined threshold values. To monitor the evolution of degradation
decrease in their overall reliability, imposing a substantial threat
due to various mechanisms, a system may be subjected to intermit-
to public safety and economic development. Moreover, capital
tent inservice inspections or online health monitoring, or both. The
and other resource constraints mean that replacement at the end
acquired data, collectively known as degradation data, serve as a
of their design life is either economically unjustifiable, or, socially
basis for degradation modeling and subsequent prognosis. While
unacceptable due to failure risks. To increase the reliability and
the degradation modeling focuses on the selection, calibration,
avoid catastrophic failures, proactive maintenance strategies based
and validation of competing models against the acquired degrada-
on the monitored health of system is desirable. Traditionally, relia-
tion data, prognosis is the systematic framework for assessing the
bility assessment and failure prediction are achieved using scarce
future condition or performances of an infrastructure asset, given
lifetime or failure data, which makes the reliability engineering
its present health condition. This is often done by projecting the
and system safety a “small data” problem. Due to recent advance-
degradation model to a predefined threshold. The end result of prog-
ments in wireless network, remote sensing, and edge computing
nosis is given in terms of remaining useful life (RUL), end-life,
technologies, it has become clear that the fourth industrial revolu-
health index, or the probability of failure at a given time in the
tion is taking place and the internet of things (IoT) era is fast
future.
approaching.
At the core of prognosis lies degradation modeling, which is a
way to mathematically quantify the underlying degradation mecha-
nism and subsequently predict the end-life of an unit, or the extent
1.1 General Framework of Damage Prognosis. Failures in of degradation of the unit at a specified future time [1]. The accu-
most of the engineered systems result from a gradual and irrevers- racy of future predictions depend upon several factors such as
ible accumulation of damage, which occur during their operation. choice of damage sensitive features (DSFs) also called damage indi-
This wear-and-tear type of situation, often named as soft failure, cators, the type and amount of data available, the type of degrada-
tion model used, and the measurement noise. Figure 1 illustrates
Manuscript received April 26, 2020; final manuscript received October 6, 2020; relationship between the degradation modeling and the end-life
published online November 10, 2020. Assoc. Editor: Shiro Biwa. prediction. Let Yt denote the univariate DSF observed at time t.

Journal of Nondestructive Evaluation, Diagnostics MAY 2021, Vol. 4 / 021005-1

and Prognostics of Engineering Systems
Copyright © 2020 by ASME
 degradation models and various statistical and stochastic modeling
techniques were thus developed. Many excellent reviews were pub-
lished that discussed the various aspects of degradation modeling
and prognosis. The notable ones include Ref. [2] for accelerated
test data, Ref. [3] for gamma process (GP) modeling, Ref. [4] for
degradation models in reliability analysis, Ref. [5] for the estimation
of remaining useful life, Ref. [6] for industrial applications of time-
based and condition-based maintenance, Ref. [7] for stochastic
degradation models of highly reliable industrial products, Ref. [8]
for deteriorating structure and infrastructure systems under uncer-
tainty, Ref. [9] with a focus on engineering applications of degrada-
tion modeling, and Ref. [10] on condition-based maintenance
optimization for stochastically deteriorating system.

1.3 Degradation Process, Nondestructive Testing, and

Fig. 1 Degradation modeling and end-life predictions Structural Health Monitoring. In past, several physical degrada-
tion processes such as fatigue, corrosion, cracking, leakage, rupture,
pitting, and creep have been studied in literature. Researchers have
The DSF can be obtained from in-service inspections, intrusive used data obtained from nondestructive testing (NDT) and struc-
examinations, or extracted from a fusion of in-service health mon- tural health monitoring (SHM) to model such processes [11]. The
itoring sensors. In addition, degradation data often have measure- main idea of any NDT technique is to send an incident signal to
ment errors. This brings extra challenges to degradation In the material (or system) and quantify the damage based upon the
general, the end-life is predicted by projecting degradation data to distortion produced in transmitted or reflected signal. The type of
the failure threshold ηD. Note that the DSF Yt is a stochastic signal produced depends upon the interaction between the signal
process with an instantaneous probability density function f (Yt), and damage present in the material. Some of the widely used
and hence, the resulting end-life is also a random variable. The NDT techniques are as follows: acoustic emission, eddy current,
end-life of an unit is the time when the degradation level first infrared thermographic testing, leak testing, magnetic testing,
reaches the predefined failure threshold, ηD i.e., strain, radiographic testing, ultrasonic testing, and visual testing.
Broadly, the inspection using NDT techniques can be divided
T = inf{t:Yt ≥ ηD } (1) into three categories: (i) inspection of raw products such as forg-
ings, castings, and extrusions, (ii) inspection following secondary
The failure time thus defined is also called first-passage time. The processing such as machining, welding, grinding, and heat treating,
value of ηD should be chosen carefully because a smaller value of and (iii) in-service inspection for damage including cracking corro-
ηD will result in premature maintenance or removal of the system, sion, erosion/wear, and heat damage. The third type of inspection is
and a high value of ηD may cause system failure without alarm. particularly important from prognosis point of view and has been
The relationship between the degradation model and end-life pre- applied in many industries such as nuclear, mechanical, civil, and
diction in practice can be much more involved than that sketched in aircraft. For example, nuclear heat exchanger tubes are inspected
Fig. 1 for at least the following two reasons. First, the system per- using eddy current probes for any corrosion damage, storage
formance is often defined by multiple DSFs. To add more complex- tanks are monitored by robotic crawlers for any signs of wall thin-
ity, the multiple DSFs may work in parallel, series, and/or ning due to corrosion, and the pressure vessels are monitored using
parallel-series topologies to define a system failure. For example, radiography and ultrasonic testing. The NDT techniques have also
if a series system is modeled using two DSFs, Yt1 , Yt2 with their cor- been used in civil engineering practice, for example, railway track
responding thresholds η1D and η2D , respectively. The failure time can inspection, bridge inspection, pipeline inspection, and damage
be given as monitoring in masonry and concrete structure [12,13]. Several
review papers have also been published in the past related to
T = inf{t:(Yt1 ≥ η1D : ∪ :Yt2 ≥ η2D )} (2) NDT studies as, for example, Ref. [14] for wind turbines,
Second, the DSFs are not directly observable as state variables. In Ref. [15] for NDT of composite materials, Ref. [12] for concrete
that case, the inference of the DSFs from the observable variables and masonry structures, and Ref. [16] for a review of ultrasonic
require more complex system modeling and Bayesian inference. method for pressure vessel and pipeline weld inspections.
Recent developments in degradation modeling reflect the progress Reviews of NDT applications are crack growth in metals [17],
toward this direction, using data and information from multiple nuclear piping system [18], oil and gas pipelines [19], and wall thin-
sources. ning due to flow-accelerated corrosion [20]. However, due to the
offline nature of NDT methods, inspection data are often scarce,
unbalanced, and subject to measurement error or other imperfect
1.2 A Brief History of Damage Prognosis. The subject of information. The imperfect information includes censored data,
damage prognosis and life prediction did not begin with degradation truncated data, interval-valued data, incomplete event history,
modeling historically. Due to the lack of reliable degradation data, unobservable heterogeneity, etc. To address these data-related
or, other failure precursor information, the prediction of RUL used issues, many advanced statistical methods have been developed,
to be done in a very rudimentary manner. The simplest approach mostly involving the use of Bayesian techniques. Interested
was to subtract a design life by the service age to obtain the readers are referred to Ref. [21] for a detailed discussion of
RUL. This approach completely neglected the variation in the various data imperfection scenarios and the modeling techniques
actual useful life of an engineering system. Also neglected were thereof.
various uncertainties in the operating factors that affect the rate of Recognizing the limitations of the offline inspection-based
deterioration and remaining useful life. Later, reliability engineering degradation prognosis approaches, people have over the past
suggested update of the RUL by applying the survival information decade started to explore the possibility of using structural and
to the lifetime probability distribution. From around 1980s, as non- machinery health monitoring data for degradation prognosis.
destructive testing and inspection technologies got more matured, Although SHM has since its early inception been aiming to
degradation has emerged to be an important piece of information predict the end-life of structures, research in early days was
for failure prognosis and life prediction. A large number of mainly focused on damage diagnosis instead of prognosis.

021005-2 / Vol. 4, MAY 2021 Transactions of the ASME

Table 1 Some recent review papers on degradation modeling mechanism of the physical degrading process is unknown. Never-
prognosis theless, the SHM-based prognosis has its own limitations. For
example, the signals are often noisy. The relevance of signals
References Remark acquired from the SHM system can be questionable. There is a
gap between the observable variables and the true system state vari-
[8] Civil structures, emphasis on corrosion and fatigue in steel and
concrete structures
ables. This naturally underscores the need for multivariate degrada-
[26] An early review on diagnosis and prognosis for tion modeling or the state-variable description of the system.
condition-based maintenance of infrastructure A summary of recent review papers in the area of degradation
[27] A review on prognostic modeling options for remaining useful modeling and prognostics have been presented in Table 1. The
life estimation by industry reviews presented earlier are either focused on a particular type of
[30] A review of physics-based models in prognostics with approach (e.g., physics-based, data-driven or hybrid approach) or
application to gears and bearings of rotating machinery on a particular type of engineering system such as gearbox.
[29] A systematic review from data acquisition to RUL prediction However, a forward-looking review under the context of big-data
for Machinery health prognostics and artificial intelligence incorporating systems from several engi-
[31] A review on current status of machine prognostics in
condition-based maintenance
neering disciplines in an integrated manner is lacking in literature,
[32] A review of Wiener-process-based methods for degradation which is addressed in this work.
data analysis and remaining useful life estimation
[33] A review on the statistical data driven approaches for 1.4 Objective and Organization. The main objective of this
remaining useful life estimation
paper is to categorize the existing prognosis approaches and their
[34] The state of the art, challenges and opportunities for rotating
machinery prognostics application to assist researchers and practitioners in acquiring a
[35] A review of hybrid prognostics approaches for remaining clear comprehension of the subject area. The existing prognosis
useful life prediction of engineered systems approaches are divided into four categories: physics-based,
knowledge-based, data-driven, and hybrid degradation models as
shown in Fig. 2.
A concise review is presented for all approaches. In each case, a
brief mathematical introduction of the model is given followed by
Choosing effective sensor technologies and developing effective
their applications, advantages and disadvantages. The method of
diagnosis algorithms have been the thrust areas. Recently, a few
parameters estimation is also discussed. The remainder of this
attempts have been made by the researchers for damage prognosis
paper is organized as follows: first, physics-based degradation
of engineering systems utilizing the SHM data [22–25]. An early
model are reviewed. The various data-driven models are surveyed
review can be found in Ref. [26], which is focused on the applica-
next, and their advantages and disadvantages are given. These are
tion of degradation diagnosis and prognosis for condition-based
followed by the knowledge-based and hybrid models. Finally, the
maintenance models. More recent reviews include prognostic mod-
key conclusions and some possible future directions are presented.
eling options for industry [27], SHM application [28], and machin-
ery health prognosis [29].
Comparing with the traditional offline inspection methods, the
SHM-based prognosis has a number of advantages. First, the data 2 Physics-Based Models
available for the prognosis are much more abundant than the Physics-based approaches (also referred as mechanistic
offline inspection data. Second, SHM often employs multiple approaches) try to model the underlying degradation process
types of sensors, which allow the fusion of information for more (e.g., brittle fracture, creep, fatigue, wear and corrosion) through
comprehensive damage diagnosis, which improves the fidelity of understanding of the mechanism behind it and formulating the gov-
prognosis. Third, the method can be employed even if the accurate erning mathematical equations. These approaches assume that the

Fig. 2 Different prognosis approaches

Journal of Nondestructive Evaluation, Diagnostics MAY 2021, Vol. 4 / 021005-3

and Prognostics of Engineering Systems
degrading system can be represented by a physical model. The main often accompanied by finite element models which are computa-
advantage of physics-based approaches is that the physical knowl- tionally expensive to build and run for large systems.
edge of the system is used for model development, which is espe- Biondini and Frangopol [8] presented a review of deterioration
cially useful for predicting the system response to new operating modeling of structural systems under uncertainty, with emphasis
and loading conditions. These models, if validated, would provide on the effects of corrosion and fatigue in steel structures and
accurate results compared to other approaches such as the data- chloride-induced corrosion in concrete structures. The corrosion
driven ones. Moreover, if understanding of the system degradation rate in reinforcing steel is a diffusion process and can be described
improves, the model can be adapted to increase its accuracy and to by the Fick’s laws, which can be reduced to simple second-order
address subtle performance problems. However, physics-based linear partial differential equation under certain assumptions. A
approaches are typically more computationally intensive than the numerical solution of the general Fick’s diffusion laws is presented
data-driven techniques, particularly, if finite element modeling of in Ref. [40] for life-cycle assessment of concrete structures exposed
a complex system is required. Finally, the validation of physics- to corrosion.
based model is a challenging task.
To develop an empirical physics-based model, several assump-
tions and simplifications are often made. Hence in practical situa- 2.1 Example: Physics-Based Models. A widely studied
tions, such a model is unable to capture the stochastic degradation problem in literature is the capacity of a secondary
characteristics of defect growth. Therefore, it is important to keep cell such as lithium-ion battery. In general, the capacity of
assumptions and simplifications realistic because real-life systems lithium-ion battery decreases as the number of charging and dis-
and infrastructures are complex and damage propagation is proba- charging cycles increases and often assumed to be failed when its
bilistic in nature. Luo et al. [36] proposes a general framework capacity reduces by 30% of the rated value. The mechanism of
for degradation modeling and prognosis using physics-based battery degradation is very complex and not well understood, and
approaches as illustrated in Fig. 3. According to them, the model- hence, an empirical degradation model (see Eq. 4) is often consid-
based prognostics process can be divided into six steps. In the ered as a physical model in physics-based approaches [41].
first step, system degradation is modeled using a set of variables
and a degradation function [37,38]. In the subsequent steps, evolu- y = a exp(−bt) (4)
tion of system dynamics is obtained through Monte Carlo simula- where a and b are model parameters, t is the time or cycles, and y is
tions, and degradation is tracked using an interacting model. the internal performance of battery, which is represented by the
Finally, the remaining life is estimated based on the current capacity at a normally rated current C1A in this numerical
damage state as well as the future usage of the system. These prog- example. The constant a is the initial degradation level and is
nosis steps are demonstrated using a simulation study of an automo- assumed to be unity in case no damage in the beginning.
tive suspension system, which is subjected to irregular excitation With the following true model parameters a = 1, btrue = 0.004, the
from the road surface. capacity data C1A were generated at every five charging and dis-
One of the most common physics-based models is the Paris crack charging cycles. Furthermore, a Gaussian noise N(μ = 0, σ 2 =
growth law [39], which is based on fracture mechanics and can 0.012) was added to this data to simulate the observed battery degra-
provide information about crack size, crack growth rate, and dation as given in Table 2.
end-life of an operational unit. According to this model, the The goal of prognostics is to estimate the parameter b and RUL
fatigue crack growth rate da/dN is related to the Irwin stress inten- by using this dataset. The method of nonlinear least square regres-
sity factor ΔK (= Kmax − Kmin) and given by the following equation sion was used, and the parameters were estimated. A histogram of
[37,39]: estimated parameters are presented in Fig. 4.
Finally, the estimated parameters were used to predict the future
da/dN = C(ΔK)m (3) degradation level and RUL, which shown in Figs. 5 and 6,
respectively.
where C and m are material- and environment-specific constants, a
The true degradation path (based upon true parameters) is also
is the crack size, and N is the number of fatigue-cycle. Although
shown in Fig. 5. It can be seen from the figure that the actual
these degradation models have shown to work well, they are
failure time is 118 cycles, i.e., actual RUL is 118 − 50 = 68
cycles. The different percentile of RUL is shown in Fig. 6, where
it can be seen that the 95 percentile of RUL is 40.85 cycles.
The estimated 95% RUL as predicted from the physics-based
model can further improve as more data are included in the model.
Despite the aforementioned example of usage, it is not always
possible to build a mathematical model for a physical system, par-
ticularly, when the degradation mechanisms are not revealed, or,
complex. Hence, the uses of physics-based prognostics are
limited. For such cases, knowledge-based and data-driven
approaches that does not depend upon a physical model appears
to be more promising.

3 Data-Driven Models
In general, the degradation signals recorded, even from the
similar systems at same environmental and test conditions, show
significant variability in their paths. The variation from one path
to another are usually considered to be uncertainties due to measure-
ment errors, unit-to-unit differences, and internal deterioration
mechanisms [42]. Purely deterministic physics-based models
cannot always explain such uncertainties, thereby presenting diffi-
culties in estimating remaining useful life. However, data-driven
Fig. 3 Physics-based approach for degradation modeling and models do not rely on physical models and use statistical principles
prognosis (adapted from Ref. [36]) in conjunction with monitored data to construct models (such as,

021005-4 / Vol. 4, MAY 2021 Transactions of the ASME

 Table 2 Battery degradation data

Inspection number 1 2 3 4 5 6 7 8 9 10 11

Time (cycles) 0 5 10 15 20 25 30 35 40 45 50
y (C1A), Amp · h 0.99 0.97 0.95 0.95 0.92 0.90 0.87 0.86 0.85 0.83 0.82

Fig. 4 Estimated parameters: (a) parameter b and (b) parameter s

Fig. 5 Future degradation level prediction using physics-based model

stochastic and bivariate process models) and account for the varia- 3.1 Cumulative Damage Damage Modeling Framework.
bility in the degradation paths. The CD assumes that the infrastructure deterioration is caused by
A diagram showing the classification of data-driven degradation multiple “shocks,” and that the damage caused by each shock accu-
model is presented in Fig. 2. In the following subsections, a review mulates over time. Let Di denotes the size of the damage caused by
of cumulative damage (CD) modeling framework, directly observ- the ith shock, N(t) the number of shocks occurred till time t, and Wi
able modeling approaches, and bivariate process modeling the sequence of inter-arrival times between successive shocks.
approaches are presented. Then, the cumulative damage Y(t) can be given by

Journal of Nondestructive Evaluation, Diagnostics MAY 2021, Vol. 4 / 021005-5

and Prognostics of Engineering Systems
 Fig. 6 RUL using physics-based model

exponential and a general distribution, respectively, then the cumu-

N(t)
Y(t) = Di (5) lative damage is given by compound Poisson process.
i=1 The CD model was first applied for the fatigue life prediction of
ball bearings by Palmgren in 1924 [44]. Over the past decades, the
A failure occurs when cumulative damage first reaches the prede- CD models have been widely applied in various fields such as com-
fined threshold ηD. From the above definition, it can be noted that posite laminates [45], fatigue life prediction [46], fiber-reinforced
the degradation Y(t) is governed by two probability laws: one for plastics [47], and railway wheel damage [48].
the timing of occurrences of the shocks, which is described by a Generally, random shocks are independent, but can often initiate
counting process N(t), or equivalently inter-arrival time model Wi, a damage or can change degradation model parameters such as,
and another for the damage size caused by the ith shock Di. The degradation rate. Hence, it is important to consider the interaction
various possibilities for random variables Di and Wi, and the corre- between random shocks and degradation processes. Rafiee et al.
sponding cumulative degradation models are presented in Fig. 7. It [49] proposed a reliability model for devices subject to dependent
can be seen from the table in Fig. 7, when Wi is fixed, say Wi = 1, competing failure processes of degradation and random shocks
and Di follows a discrete distribution then Yi follows the discrete- with a changing degradation rate according to particular random
time Markov chain model. Similarly, if Wi and Di follows shock patterns. For complex multi-component systems with each

Fig. 7 Various cumulative degradation models (adapted from Ref. [43])

021005-6 / Vol. 4, MAY 2021 Transactions of the ASME

component experiencing multiple failure processes due to simulta- 3.2.2 Wiener Process. Wiener Process belongs to a special
neous exposure to degradation and shock loads, a multi-component class of stochastic processes with stationary and independent incre-
system reliability model was developed by Song et al. [50]. Wang ments called Lévy process, which have been used extensively in
and Pham [51] proposed a dependent competing risk model for a degradation modeling [42]. A stochastic process Y(t) is said to
deteriorating system subject to shocks. They assumed that shocks have stationary and independent increments, if it has the following
follow a non-homogeneous Poisson process with an increasing two properties: (i) for any 0 ≤ t1 ≤ · · · ≤ tn, random variables Y(t1) −
rate of occurrence based on the fact that an aged system is more vul- Y(0), Y(t2) − Y(t1), …, Y(tn) − Y(tn−1) are independent and (ii) for
nerable to the shocks. In recent times, attempts have been made to any t ≥ 0 and △t > 0, the distribution of increments ΔY(t) = Y(t +
incorporate both gradual and shock deterioration processes into the Δt) − Y(t) depends on the time difference Δt.
life-cycle analysis of the engineering systems [52]. In general, a random increment in damage ΔY(t) can be treated as
the sum of large number of small degradations caused by indepen-
dently and identically distributed random external factors. Using the
3.2 Directly Observable Modeling Approaches. For most central limit theorem, it can be concluded that the every increment
civil infrastructure systems and other engineered products, degrada- ΔY(t) is normally distributed. This is exactly one of the properties of
tion is a monotonic process, where the damages occur sporadically the Weiner process and makes the Wiener process a well-suited can-
in time with damage size also distributed randomly. Therefore, most didate for degradation modeling [58]. A generalized Weiner process
of the degradation phenomena can be modeled using stochastic can be obtained by introducing drift parameter (ν) and diffusion
models such as Markov and semi-Markov models, Wiener parameter (σ) into the standard Wiener process. According to the
process, gamma process, inverse Gaussian (IG) process and multi- generalized Weiner process model, the cumulative degradation
variate process models. Y(t) at any time t follows normal distribution with mean νΛ(t)
and variance σ 2Λ(t), where Λ(t) is the scaled time, and it is a con-
tinuous monotonically increasing function. Let B(·) be a standard
3.2.1 Markov and Semi-Markov Models. A stochastic process Brownian motion, then the degradation path in a Wiener process
in which future states are conditionally independent of their past can be written as follows:
states, given the present state, is known as Markov chain. Mathe- Y(t) = νΛ(t) + σB(Λ(t)) (9)
matically, it can be written as follows:
The coefficient of variation (COV) for this process can be given by
Pr(Sn+1 = j|Sn = i, . . . , S1 = i1 , S0 = i0 )
= Pr(Sn+1 = j|Sn = i)
(6) Var(D(t)) σ
COV(Y(t)) = = √ (10)
E(D(t)) ν Λ(t)
where Pr(·) denotes probability. In the past, engineering judgment
has been widely used to describe the system’s state using condition It can be seen from Eq. (10) that as time increases, the coefficient of
ratings. Some examples are the pavement condition index [53] and variation decreases, and consequently, the sample degradation path
the bridge deck condition rating [54]. Let Pij denotes the probability Y(t) approaches asymptotically to the mean path.
of transition of system from the ith state to jth state, pi is the prob- The probability density function (PDF) of damage increments
ability that the system is in ith state at t = 0, then a Markov chain can ΔY(t) can be derived using the property of the Weiner process,
be characterized by one-step transition probability matrix P and i.e., ΔY(t) follows a normal distribution with mean νΔΛ(t) =
initial state probability vector p 0 as follows: νΛ(t + Δt) − νΛ(t) and variance σ 2ΔΛ(t) = σ 2Λ(t + Δt) − σ 2Λ(t).
The PDF of ΔY(t) can be given as follows:
P = [Pij ] where Pij = Pr(Sn+1 = j|Sn = i) (7)  
1 (Δy − νΔΛ(t))2
f (Δy; ΔΛ(t), ν, σ) = √ exp − (11)
p0 = [pi ] where pi = Pr(S0 = i) (8) σ 2πΔΛ(t) 2σ 2 ΔΛ(t)

In general, the transition probability matrix is either estimated Furthermore, it can be shown that the failure time T follows a trans-
using the field data, or, based on engineering judgment. Madanat formed inverse Gaussian distribution, i.e., Λ(T) ∼ IG(ηD /ν, η2D /σ 2 ),
et al. [54] proposed a method for the estimation of infrastructure and the CDF of Λ(T ) is given by
transition probabilities from condition rating data. Estimation of  
transition probabilities for bridge deterioration and subsequent fore- η2D νt
FΛ(T) (t) = Φ − 1
casting is proposed in Ref. [55] using multiple exponential hazard σ 2 t ηD
models. The main assumptions of Markov model are as follows:    
2νηD η2D νt
(i) Transition probability does not change with time. + exp Φ − + 1
(ii) The sojourn time in a particular state is exponentially σ2 σ 2 t ηD
distributed. ηD
(iii) The sum of all transition probabilities out of each state into The expectation and variance of Λ(T ) are E(Λ(T)) = and
another different state must be equal to one. ν
Var(Λ(T)) = ηD σ 2 /ν3 , respectively.
The assumption that the sojourn time is exponentially distributed Wiener process and its variants such as those with time scale
is impractical for many of applications, and hence, Markov model is transformation, with measurement error, and with a drift have
not an appropriate choice in those cases. This issue can be addressed been widely applied in the literature for degradation modeling.
by using the semi-Markov model, where the time spent in a partic- Doksum and Hbyland [59] modeled the variable-stress accelerated
ular state can be given by any distribution. A hybrid degradation- life testing fatigue failure data using a WP. In their study, the accu-
based reliability model for a single-unit system whose degradation mulated fatigue decay was modeled using two separate WPs, which
is driven by a semi-Markov environment was proposed by change from one to another at a certain stress change point, t. Wang
Ref. [56]. In a recent study, semi-Markov model is used for the oxi- [60] applied a WP with random effects to analyze bridge beam
dation degradation modeling of gas turbine nozzles [57]. Despite degradation data. In their model, the unit-to-unit variability was
the use of Markov and semi-Markov model, the output of the incorporated through random effects, and the uncertainties in the
process is its state, which corresponds directly to a physical, observ- model parameters were estimated using a bootstrap method. In
able event. Nonetheless, in many practical applications, the hidden another study, Si et al. [61] used a WP-based degradation model
states of the system can only be inferred through indirect sensor with a recursive filter for RUL estimation of gyros in an inertial nav-
measurements such as vibration, strain, and displacement. igation system. They were able to update the drift coefficient of WP

Journal of Nondestructive Evaluation, Diagnostics MAY 2021, Vol. 4 / 021005-7

and Prognostics of Engineering Systems
using a recursive filter and other parameters using an EM algorithm. Moreover, the model accounts for the temporal uncertainty in the
In this way, the updating was done in real-time upon the availability degradation process. However, its Markov property and the fact
of on-line degradation data. that it is strictly applicable to monotonic processes may restrict its
Time scale transformation of a Wiener diffusion process was con- application for some degradation processes.
sidered by Whitmore and Schenkelberg [62] and applied toward
degradation modeling of a self-regulating heating cable. The 3.2.4 Inverse Gaussian Process. Another Lévy process that
problem of measurement noise in the degradation signal, which provides a monotone degradation path is IG process, which is
gives an inaccurate WP parameter estimates was undertaken by well suited for the cases, where damage increments ΔY(t) follows
Ye et al. [63]. They modeled a traditional WP with positive drifts IG distribution. In contrast to the application of WP and GP, the
compounded with identically and independently distributed Gauss- IG process is relatively less applied in the past [73]. Perhaps, this
ian noises and improved its estimation efficiency when compared is due to the lack of physical understanding of this process.
with existing inference procedures. They developed methodologies Recently, Ye and Chen [74] tried to give a physical meaning of
that are applicable to the wear problem of hard disk drive magnetic this process. They also found that similar to GP, the IG process is
heads and the light intensity degradation problem of light-emitting a limiting case of CPP, but the shock size distribution is different
diode lights. WP is good for modeling the stochastic nature of from that of GP. According to the definition of the IG process,
degradation, which can increase or decrease between two consecu- the damage increment ΔY(t) follows IG distribution, i.e., ΔY(t) ∼
tive times. However, in many cases, degradation is monotonically IG(μΔΛ(t), λΔΛ2(t)), where ΔΛ(t) = Λ(t + Δt) − Λ(t) and Λ(t) is
increasing, where this cannot be applied. a monotonically increasing function and μ, λ > 0. The functional
forms of Λ(t) can be linear, power law or exponential, which is
3.2.3 Gamma Process. The GP is a Lévy process with non- often chosen based on particular engineering applications. More-
negative increments, which follows a gamma distribution with an over, the PDF of cumulative degradation also follows IG distribu-
identical scale parameter. The GP is an appropriate choice when tion, i.e., Y(t) ∼ IG(μΛ(t), λΛ2(t)) and given by
the degradation path is monotonically increasing and random incre-   
ments ΔY(t) follows gamma distribution. For degradation path fol- λΛ2 (t) λ(y − μΛ2 (t))
lowing GP, the PDF of damage increment ΔY(t) is given by f (y; μΛt, λΛ (t)) =
2
exp − (16)
2πy3 2μ2 y
uΔv(t)
f (Δy; Δv(t), u) = exp {−uΔy} (12) The CDF of failure time distribution can be obtained by considering
Γ(Δv(t)) its monotonicity property and distribution characteristics of degra-
∞ dation increments. Mathematically, it can be written as follows [73]:
where Γ(·) = z=0 za−1 e−z dz is the gamma function for a > 0, Δv(t) =
v(t + Δt) − v(t) is the change in shape function, and u is the scale
FT (t) = Pr{T < t} = Pr{Y(t) > ηD }
parameter. Furthermore, by using the additive property of gamma (17)
distribution, the PDF of cumulative degradation Y(t) till time t is = 1 − FIG (ηD ; μΛ(t), λΛ2 (t))
written as follows:
    
uv(t) λ ηD 2λΛ(t)
f (y; v(t), u) = exp {−uy} (13) =1−Φ − Λ(t) − exp Φ
Γ(v(t)) ηD μ μ
The CDF of lifetime distribution can be written as follows:   
λ ηD
× − + Λ(t) (18)
FT (t) = Pr{T < t} = Pr{Y(t) > ηD } (14) ηD μ

∞ ∞
uv(t)
= f (y)dy = exp {−uy}dy where G(.; μΛ(t), λΛ2(t)) is the CDF of IG(μΛ(t), λΛ2(t)) and Φ(·)
ηD ηD Γ(v(t)) are standard normal CDF. Thus, for IG process there is a one-to-one
(15)
Γ(v(t), ηD u) relationship between Λ(t) and the CDF of the failure time distribu-
= tion FT(t) for fixed parameter λ and μ. Moreover, the IG process can
Γ(v(t))
easily be extended to incorporate random effects (i.e., unit-to-unit
∞
where Γ(a, y) = z=x za−1 e−z dz is the incomplete gamma function. variability) and covariates such as temperature, voltage, and pres-
The parameters of GP model can be estimated using maximum sure in the IG process model. Wang and Xu [73] applied IG
likelihood, method of moments, and Bayesian statistics [64]. A process to the laser data where both the unit-to-unit heterogeneity
Bayesian approach of parameter estimation and its update were pro- and covariate information were incorporated into the model. In
posed by Guo and Tan [65]. Ye et al. [66] obtained maximum like- their study, the EM algorithm was used to obtain the maximum like-
lihood estimates of the parameters using the EM algorithm. In their lihood estimates of the unknown parameters and bootstraping is
method, the bootstrap technique is used to construct confidence used to assess the variability in estimates. Recently, Peng et al.
intervals. [75] conducted a Bayesian analysis of inverse Gaussian process
The various extension of GP such as multivariate GP, weighted models for degradation modeling and inference. Apart from a
GP, and non-stationary GP are also employed for degradation mod- simple IG process model, they also investigated three IG process
eling. van Noortwijk [64] presents an exhaustive survey of models with random effects using the Bayesian method.
GP-based degradation modeling for civil engineering infrastructure.
A stochastic GP model was used for building deterioration by Edir- 3.2.5 Time Series Models. Time series models such as auto-
isinghe et al. [67], for creep of concrete by Cinlar et al. [68], for regressive (AR), auto-regressive moving average (ARMA), and
fatigue crack growth by Lawless and Crowder [69], and for thinning auto-regressive integrated moving average (ARIMA) models have
caused by corrosion by van Noortwijk et al. [70]. Grall et al. [71] been widely used for short-term forecast [76]. The fundamental
used a GP to analytically model the deterioration of a continuously idea of these models is to represent the future DSF values as a
deteriorating single-unit system and proposed an optimum preven- linear function of past DSF observations and random errors. In
tive replacement policy structure. Liao et al. [72] presented a general, an ARMA(p,q) model consists of p autoregressive (AR)
condition-based maintenance model for continuously degrading and q moving average (MA) parameters and can be written as
systems under continuous monitoring. follows [76]:
In past, GP has been widely used for degradation modeling due to
its physical interpretation and rich mathematical structure. ϕ(B)(Yt − μ) = θ(B)at (19)

021005-8 / Vol. 4, MAY 2021 Transactions of the ASME

where ϕ(B) = 1 − ϕ1B − ϕ2B 2 − · · · − ϕpB p is the autoregressive particle filtering methods and Bayesian Networks (BNs). Weidl
operator of order p and θ(B) = 1 − θ1B − θ2B 2 − · · · − θqB q is the et al. [87] provided a guide for selecting a suitable technique for
moving average operator of order q, at is a white noise process a given type of system and noise. These techniques utilize SHM
with zero mean and variance σ 2a . The backward shift operator B is observations obtained from the component/system of interest to
defined by Byt = yt−1 and B kyt = yt−k. Note that when q = 0, the monitor the damage progression. Predictions of one damage state
ARMA model reduces to a AR model. The parameters ϕi (i = 1, to another state involve integrals that cannot always be evaluated
2, …p), θj ( j = 1, 2, …q), and σ 2a of this model are estimated in closed form. Thus, integration approximation methods are
using the Yule-Walker method [76]. The model order (p, q) is often required, such as regression models [88] or bootstrapping
selected based on Akaike information criterion (AIC) or Bayesian methods [89], to estimate the expected value and covariance of
information criterion (BIC). In summary, the development of an these probability density functions.
ARMA model is done in the following three steps: (i) model iden-
tification, (ii) parameter estimation, and (iii) model validation. 3.3.1 Bayesian Networks. BNs are probabilistic acyclic graph-
Wu et al. [77] proposed an ARIMA approach for prognosis of ical models that represent a set of random variables and their prob-
rotating machinery using vibration data collected from the abilistic interdependencies. Such a network can be built using
machine. Yan et al. [78] employed an ARMA time series model information obtained from domain knowledge, SHM data, or
to trend the probability of failure calculated by a logistic regression both. A Bayesian network consists of nodes, which represent
model to estimate the RUL. In addition, to express change of var- random variables that can take on distinct states or levels. These
iance of time series data over time, a GARCH (generalized autore- are connected by directional arcs that represent a direct causal influ-
gressive conditional heteroscedastic) model is used in Kobayashi ence between nodes in a mandatory acyclic pattern; conditional
et al. [79]. Time series models are effective for short-term predic- probabilities are used to quantify the strengths of these causal influ-
tions, but less reliable when used for long-term predictions due to ences. The result is that each node has a conditional probability
dynamic noise, their sensitivity to initial system conditions and an table that defines probabilities for each state of the node given the
accumulation of systematic errors in the predictor [76]. For long- states of its parents [90]. There have been few applications of
term predictions, long memory models would be useful, and Koba- BNs to prognostics. An adaptive Bayesian decision model was pro-
yashi et al. [80] proposed methodology to use an autoregressive posed by Kallen and van Noortwijk [91] to determine the optimal
fractional integrated moving average-fractional integrated general- periodic inspection and replacement policy by minimizing the
ized autoregressive conditional heteroscedastic model, which is expected cost rate. Weidl et al. [87] proposed the concept of
one of long memory models, for deterioration forecasting with objected-oriented Bayesian networks (OOBNs) for process predic-
time series monitoring data. tion using one step ahead time series prediction. Recently, Yuan
et al. [92] applied BNs for probabilistic fatigue life prediction of
3.2.6 Accelerated Test Degradation Models. Accelerated test concrete bridges. In spite of BNs application for prognosis, the
(AT) degradation models make end-life predictions at normal con- accuracy of its result depends upon the choice of prior distribution,
ditions using the degradation data obtained at accelerated conditions and the reliability of prior knowledge. Therefore, some understand-
(e.g., high temperature, high stress or high voltage). An AT degra- ing of the physical system is useful. Kalman and particle filters are
dation model can be thought of as the combination of two sub- two different approaches to implement the generic dynamic Baye-
models namely, a stochastic model that describes the degradation sian networks and are reviewed in the following sections.
process and an acceleration model interpreting the functional rela-
tionship between degradation evolution and stresses [81]. This 3.3.2 Kalman Filter. The Kalman filter (KF) is a computation-
type of model is particularly applicable for slow degrading infra- ally efficient recursive statistical signal processing technique, which
structure, where the degradation data till failure are difficult to is used to estimate the state, parameters, and sometimes a combina-
obtain under normal conditions. In such cases, the degradation tion of both informations, of a dynamic system from a series of
data are collected at accelerated conditions and extrapolated incomplete and noisy measurements by minimizing the mean-
through AT model to obtain the estimates of life or long-term per- squared error. At any instant, the KF is defined by its state and
formance at lower and normal operating conditions. Lydersen and error covariance, which is estimated from the SHM data available
Rausand [82] presented a systematic approach to perform acceler- till that time. The KF algorithm assumes that the process noise
ated life testing. Park and Padgett [83] provided several accelerated and the measurement noise are Gaussian, white, independent, and
degradation models for failure based on geometric Brownian additive. The KF algorithm works best when the dynamic system
motion and gamma processes. being modeled is linear. However, extended Kalman filter
In past, AT models have been applied for various applications (EKF)—a modification of basic Kalman filter—can be applied
such as light-emitting diodes [84], induction motors [85], and even if underlying process itself or the relationship between the
rolling element bearings [86]. Escobar and Meeker [2] presented process and the measurements is nonlinear. Wang [93] developed
a review of AT models, which have been applied successfully in a framework using KF to predict the conditional residual life of
the past. The AT model can either be parametric or non-parametric. engineering system. Their model considers both the unobserved
For parametric cases, a probability distribution (e.g., Weibull) is condition, xt, and the observed SHM data, yt, as non-stationary sto-
chosen in order to estimate the reliability function. However, a non- chastic processes, such that xt = αxt−1 + ϵt and yt = βyt−1 + ηt where
parametric model does not need an assumption on probability dis- ϵt and ηt are Gaussian noises, and α and β are the parameters of the
tribution, and the reliability function is estimated using the state space model. The KF was used to predict xt+k given SHM data
Kaplan–Meier estimator. Even though the AT model is widely to date, i.e., yt = y1, y2, …, yt. In an another study, Batzel and
applied for electrical and mechanical components, its application Swanson [94] presented a RUL estimation method based on the
for civil infrastructure such as bridges, dams, and pipe corrosion Kalman filter for aircraft power generators.
is limited.
3.3.3 Particle Filter. Particle filters are often used to estimate
the posterior distribution in a Bayesian network framework when
3.3 Bivariate Process Modeling Approaches. Bivariate either linearity or Gaussian noise assumptions of Kalman filter
models describe current state of a system in the form of a condi- are not met. Particle filter uses Monte Carlo approximation
tional reliability function and then use the Bayes theorem to instead of Gaussian approximation of the filtering distribution. In
update probabilistic estimates of future behavior as the new obser- contrast to Kalman filters—where the posterior PDFs are extrapo-
vations are made. Predictions for the new state (i.e., the condition lated from the prior state, particle filters use sequential importance
probability or posterior function) can be improved by incorporating sampling (SIS) to simulate the entire next state in every iteration
more advanced state estimation techniques such as Kalman and of the filter. This is achieved by drawing a set of random samples

Journal of Nondestructive Evaluation, Diagnostics MAY 2021, Vol. 4 / 021005-9

and Prognostics of Engineering Systems
(also known as particles), from a theoretical density function and prognostics capability. The authors successfully applied HSMM
then adjusting the associated set of particle weights at each iteration. for the diagnosis and prognosis of pump failure. In recent studies,
With sufficient samples, particle filters are more accurate than Peng and Dong [105] have used age-dependent HSMMs to
extended Kalman filter. Orchard and Vachtsevanos [95] used a non- predict the health state of equipment. The major disadvantage of
linear state-space model and a particle filtering to predict the evolu- HSMMs is its computational complexity that occurs due to the
tion of the fault indicator and subsequently RUL PDF. The particle inclusion of variable state duration densities in any state.
filters have been used extensively in both econometrics and target
trajectory forecasting. However, its engineering application for 3.3.5 Machine Learning Methods. A number of degradation
remaining useful life prediction is limited. Similarly, Cadini et al. models based on machine learning and artificial intelligence techni-
[96] used particle filtering for estimating the failure probability of ques have been proposed for various types of assets in the last two
a component subject to fatigue degradation. Recently, Tang et al. decades. Gebraeel and Lawley [106] developed a neural network
[97] used a stochastic filter-based method to predict RUL under degradation model to compute and update residual life distributions
several special cases, such as the Benes filter, particle filter, and of rolling contact thrust bearings. Tran et al. [107] used a neural
multiple-model filter. network model to predict structural condition of storm-water pipe
assets in Australia. Tatari et al. [108] built an artificial neural
3.3.4 Hidden Markov Model and Hidden Semi-Markov network model to assess conditions of highway culverts and
Model. A HMM is a doubly embedded stochastic process with showed that culvert management systems might be improved by
an underlying stochastic process that is not directly observable taking advantage of artificial intelligence. Tabatabaee et al. [109]
(hidden) and can be observed only through another stochastic developed a two-stage model to predict pavement performance by
process yielding the sequence of observations [98]. It is an exten- combining a recurrent neural network model with a support
sion of Markov chains. Figure 8 illustrates the architecture of vector classifier and found that the two-stage model has better pre-
HMM, which is characterized by observation sequence O, state- diction results than the single-stage recurrent neural network model.
sequence S, state transition probability matrix A, and emission Lee et al. [110] proposed a methodology to use an Elman neural
probability matrix B. network model to predict long-term performance of bridge ele-
The main advantage of using HMM is that the specific knowl- ments. Other recent studies [111–115] proposed degradation
edge of the degradation mechanism is not required. Moreover, indi- models of bridge elements using machine learning and artificial
rectly taken sensory measurements are enough to map the intelligence techniques. In addition, sewer pipes condition predic-
degradation process. Among one of the early studies using HMM, tion models based on artificial intelligence techniques can be
Butros and Liang [99] applied discrete HMM for mechanical fault found in Ref. [116]. Some of these models that include rule-based
detection and validated the proposed technique for bearing and artificial intelligence may be considered as a hybrid method of a
cutting tools. Baruah and Chinnam [100] used HMM for diagnos- data-driven model and a knowledge-based model.
tics and prognostics of machining processes. Zhang et al. [101] pre-
sented an integrated approach to bearing fault diagnosis and 3.3.6 Change Point-Based Models. In some cases, such as,
prognostics. In their studies, features from vibration data were rolling element bearings, capacitors, and liquid coupling devices
extracted using principal component analysis (PCA), which was (LCD’s), the degradation path often contains change points (CP),
further utilized to train HMMs. Kobayashi et al. [102] used the which occur due to a sudden change in the operating or in the envi-
HMM to forecast future degradation processes correcting measure- ronmental conditions, impurities present in the material, or a
ment errors in rutting inspection data of pavements. Lethanh et al. damage initiation. For precise modeling of such paths, the CP
[103] proposed a methodology to use the HMM to express genera- needs to be incorporated in the degradation model. Figure 9 illus-
tion processes of pot holes, which varies depending on a value of a trates CP-based degradation model, where a two-phase degradation
pavement deterioration indicator, as a Poisson hidden Markov path is shown. The degradation occurring within a phase can either
model. The HMM output was the degradation index, which was be described by a stochastic process or a RV model. Hence,
subsequently used by an adaptive prognostics algorithm for calcu- CP-based model has great flexibility and can be used to model a
lating the remaining useful life in real-time. wide range of degradation phenomenon. Let λ1, λ2, …, λ(n−1)
Despite of its popularity, HMM has some shortcomings such as denote the change point locations and Y0 , Yλ1 , Yλ2 , . . . , Yλ(n−1) be
the number of hidden states are not known a priori, and the method the degradation at the beginning of the 1st, 2nd, 3rd, …, nth
is computationally expensive–even for the simplest models with phases, respectively.
few states. Furthermore, Markovian assumption that the sojourn
time in any state is exponentially distributed is unrealistic in some
practical applications. To overcome this issues, Dong and He
[104] first pointed out the use of hidden semi-Markov model
(HSMM) for degradation modeling.
An HSMM is a HMM with temporal structure, which allows
modeling the time duration of hidden states and provides the

Fig. 8 Architecture of a Hidden Markov model Fig. 9 Schematic of a two-phase model with a change-point

021005-10 / Vol. 4, MAY 2021 Transactions of the ASME

 Fig. 10 Future degradation level prediction using data-driven model

Then, a general multi-phase degradation model can be written as A two-phase degradation modeling approach was used for rolling
follows: elements bearings by Gebraeel [117]. In the aforementioned work,
the authors developed a two-phase degradation model using the
Yt = [Y0 + Y1 (t)]I(0,λ1 ) (t) + [Yλ1 + Y2 (t − λ1 )]I(λ1 ,λ2 ) (t) root-mean-square value of acceleration, and they integrated real-time
(20)
+ [Yλ(n−1) + Yn (t − λ(n−1) )]Iλ(n−1,∞) ) (t) condition data in a Bayesian framework. In their model, the first
phase consists of a relatively constant signal representing the
where I(a,b)(t) is the indicator function, which is unity in interval [a, period before damage initiation, followed by a second phase—
b] and zero elsewhere. The simplest case of a two-phase degrada- monotonically increasing—representing damage progression.
tion model can be written as follows: Using this approach, they arrived at a closed-form solution to
predict and update the RUL. Change point detection along with
Yt = [Y0 + Y1 (t)]I(0,λ) (t) + [Yλ + Y2 (t − λ)]I(λ,∞) (t) (21) RUL estimation was studied by Chen and Tsui [118] and Prakash
For example, if the functional form is linear in both phases, the et al. [119]. The authors in these studies proposed a Bayesian
model can be written as follows: change point detection approach for two-phase degradation model-
 ing and RUL predictions.
θ1 + θ2 t + ϵ1 if t ≤ λ Two-phase degradation modeling with two different functional
Yt = (22)
β1 + β2 t + ϵ2 if t > λ forms in the two phases has been undertaken by Wang et al. [120]
for real-time reliability evaluation of LCDs, in which the initial
where λ is the change point location, ϵ1, ϵ2 are the additive errors, sharply increasing phase was modeled using a Gamma process,
which are assumed to be i.i.d and normally distributed with mean while the latter using a Weiner process. Feng et al. [121] used a multi-
0 and variances σ 2 and τ 2, respectively. Let θ1 = {θ1, θ2, σ 2} and phase Weiner degradation model for storage life prediction for a
θ2 = {β1, β2, τ 2} then the model parameters Θ = {θ1, θ2, λ} can high-performance capacitor. It was found in this study that ignoring
be estimated either using a maximum likelihood or Bayesian change points that exist in the degradation path could lead to signifi-
approach. According to the Bayes theorem, the posterior distribu- cant prediction uncertainty in the storage life of a capacitor.
tion of the parameters, p(Θ|Y) can be given as follows:
π(Θ)L(Y|Θ) 3.4 Example: Data-Driven Models. The problem of battery
p(Θ|Y) = ∝ π(Θ)L(Y|Θ) (23)
p(Θ)L(Y|Θ)dΘ degradation as discussed in Sec. 2.1 is further undertaken using a
data-driven model. A regression-based model, namely, Gaussian
where π(Θ) is prior distribution obtained from the past degradation process model has arbitrarily been chosen for this purpose. The
and L(Y|Θ) is the likelihood of observed data Y = {Y1, Y2, …, Yt}. Gaussian process regression (GPR) has widely been used in data-
For the degradation model in Eq. (22), the likelihood function driven prognostics. In contrast to linear regression, which
L(Y|Θ) is written as follows: assumes that the error terms are independent and identically distrib-
  uted (i.i.d.), the GPR assumes correlated errors and results in rela-
L(Y|Θ) = ϕ(Yt ; θ1 + θ2 t, σ 2 ) ϕ(Yt ; β1 + β2 t, τ2 ) (24)
t≤λ t>λ
tively accurate predictions.
The results of GPR simulations for battery degradation are shown
where ϕ(Yt; θ1 + θ2 t, σ ) denotes a normal probability density func-
2
in Figs. 10 and 11, respectively.
tion with mean θ1 + θ2 t, variance σ 2 and evaluated at Yt. The main Figure 10 shows the predicted median and 90% prediction inter-
assumptions of change point model is as follows: (i) the condition val similar to physics-based model. The threshold that is to 70% of
of the infrastructure deteriorates with operating time and the level initial C1A value and true degradation path are also shown in the
of deterioration can be observed at periodic intervals; (ii) the infra- figure. It can be seen from the figure that the predicted degradation
structure being monitored comes from a population of similar units, path is well in agreement with the actual degradation path. Finally, a
and each of which exhibits the same degradation form; (iii) the distri- histogram of predicted RUL at 50 cycles is plotted in Fig. 11.
bution of the random term across the population of devices is known; Note that, the actual RUL for the battery under consideration is
and (iv) a change point occurs only due to the damage. 68 cycles, while the 95% prediction is 36.2 cycles. The accuracy

Journal of Nondestructive Evaluation, Diagnostics MAY 2021, Vol. 4 / 021005-11

and Prognostics of Engineering Systems
 Fig. 11 RUL of battery using data-driven method

of prediction will further increase as the more data will be included (e.g., failure time is less <2 year) when such rules are known to
in this model. the human experts. In formulation of expert systems, a statement
can be either true or false; data can thus be included or excluded
from a set. However, it is not always feasible to define sets and asso-
4 Knowledge-Based Models ciated membership so precisely. Fuzzy logic is one of the first exten-
sions of expert system in which a probability is associated with each
Knowledge-based approaches are generally employed when
rule in place of simply using true or false rules to represent the
accurate mathematical models are difficult to build due to real-
knowledge.
world complexity or due to various limitations of using model-
based approaches. A knowledge-based system is essentially
composed of two sub-systems: the knowledge base and the infer- 4.2 Fuzzy Logic. Fuzzy logic is a problem-solving methodol-
ence engine. Two typical examples of knowledge-based systems ogy that provides a simple way to arrive at a definite conclusion
are expert system and fuzzy logic. based upon vague, ambiguous, imprecise, noisy, or missing input
information. Fuzzy logic is widely used in various systems ranging
from simple, small, embedded micro-controllers to large, networked
4.1 Expert System. An expert system is a computer that emu-
PC or workstation-based data acquisition and control systems. In
lates the decision-making ability of a human expert [122]. It gener-
other words, fuzzy set theory allows a partial set membership
ally consists of a knowledge base containing accumulated
based on a variable’s “degree of truth.” As in expert systems, fuzzy
experience from the experts in the field and a rule base for applying
logic systems use simple, empirically derived, IF-THEN rules to
that knowledge to particular problems known to the software
solve problems. These rules are descriptive but unlike expert
system. In other words, expert systems are designed to solve
systems, they are intentionally imprecise. A typical fuzzy process
complex problems by reasoning through bodies of knowledge,
logic statement may look like “IF (process is too hot) AND
represented mainly as IF-THEN statements rather than through con-
(process is heating rapidly) THEN (cool the process quickly).”
ventional procedural code. The process of building an expert system
The fuzzy logic-based method has been successfully imple-
involves knowledge acquisition, knowledge representation, and the
mented for several applications such as dam health monitoring
verification and validation of models [122]. Once an expert system
[124], prediction of end-life of boiler tubes [125], active control
is built, it can be used for solving problems which are normally
of bridge structures [126], and active control of high rise buildings
solved by human specialist. The output of an expert system is gen-
[127]. Interested readers for the several engineering applications of
erally understandable by non-specialist users, and logic for a partic-
fuzzy logic are directed to the book by Ross [128]. In the following
ular result can be established. Expert system has been widely
section, a numerical example is presented to illustrate the
applied for various purposes, such as for diagnosis, prognosis, inter-
knowledge-based model.
preting, monitoring industrial systems, and predictive maintenance
activities [31]. Liao [123] presented an exhaustive review of appli-
cations of expert systems between 1995 and 2004. 4.3 Example: Knowledge-Based Model. A numerical
Despite of its wide applications, the expert system approach has example using Fuzzy logic to estimate the residual life is presented
some drawbacks. First, the problem of “combinatorial explosion,” here. The details of the algorithm can be found in Ref. [123]. Let us
which often occurs when the number of rules (i.e., if-then state- consider the data given in Table 3 for two sample units of an infra-
ment) increases dramatically and causes computational issues. structure, which includes observed degradation index (x1), thresh-
Second, the performance of an expert system depends upon the old (xs), observed degradation rate per year (x2), and elapsed time
experts that develops the laws of reasoning. Finally, it is not possi- (y). Using this dataset, it is desired to obtain the residual life of
ble to make continuous variable predictions using expert system infrastructure facility.
because the output is determined by a discrete set of rules. This In Table 3, each sample is observed data from different elements
limits their ability to estimate remaining useful life. However, an of the facility. In fuzzy logic, the membership function is defined to
expert system can provide an approximate output in general terms represent practitioner’s knowledge, and here, it is assumed

021005-12 / Vol. 4, MAY 2021 Transactions of the ASME

 Table 3 Input data for knowledge-based example life of the facility. In this case, the values of μ(x, y) are compared
between the two samples, and a sample whose μ(x, y) is larger
Sample 1 Sample 2 than the other is selected. In Fig. 12, the value of the membership
function and estimated residual life are shown depending on the
Observed degradation index: x1 1 2.5 weight α. Residual life is estimated as 6.7 years if Sample 1 is
Threshold: xs 3 3
Observed degradation rate per year: x2 0.3 0.05
selected, and it is estimated as 10 years if Sample 2 is selected. In
Elapsed time (year): y 5 15 the result, when α is less than 0.75, Sample 2 is selected and residual
life of the facility is estimated as 10 years. In the same manner,
when α is greater than 0.75, residual life is estimated 6.7 years.
As it can be found in those results, fuzzy logic (a knowledge-based
as follows: model) reflects empirical knowledge (α in this case) to estimate resi-
dual life of infrastructure facilities.
β−x y
μ(x, y) = α + (1 − α) (25)
β β
5 Hybrid Models
where x = (xs − x1)/x2 is residual life calculated by observed data, α In many cases, it may be simplistic to label the degradation models
is a weight between empirical life expectancy and residual life under the umbrella of lifetime-based, knowledge-based, physics-
based on observed samples, and β is the empirical life expectancy. based, or data-driven approaches. However, for some practical appli-
α and β are exogenously given to express empirical knowledge. cations, it is necessary to hybridize either of these models. For
Residual life of an observed sample, which has the maximum example, the development of cracks in concrete under fatigue, stiff-
value of the membership function, is used as the estimated residual ness degradation actuated by operational changes and external

Fig. 12 Result of numerical example (a) values of the membership function and (b) estimated remaining useful life

Fig. 13 Installed Instruments in TK dam (a) strain Gauge and (b) seismic tri-axial MEMS accelerometer and recorder

Journal of Nondestructive Evaluation, Diagnostics MAY 2021, Vol. 4 / 021005-13

and Prognostics of Engineering Systems
 (a) (b)
Fig. 14 Strain gauge reading (a) left block of crest and (b) right block of crest

dynamic loads such as earthquakes and hurricanes cannot be HM4: data-driven model + physics-based model; and (v) HM5:
completely described using physics-based models or purely stochas- experience-based model + data-driven model + physics-based model.
tic processes. Similarly, remaining useful life of a mechanical com- In HM1 approaches, the domain knowledge is integrated with
ponent such as a cutting tool or rolling element bearing cannot be continuous SHM data to build a model. The domain knowledge pro-
accurately predicted only by using a data-driven techniques [129]. vides an insight into the system’s fault states, while data-driven
In such cases, a hybrid degradation model is particularly useful, models are used to refine the rules generated from expert knowledge
where the partially understood degradation mechanism, and SHM and to perform the actual RUL prediction. Using this approach,
data acquired from the system can be utilized to build a model. Garga et al. [130] presented an automated reasoning method that
The fundamental assumption of hybrid model is that the use of uses rule-based expert system, and neural network for the remaining
various models helps to model the degradation process efficiently. useful life prediction of an industrial gear-box. The HM2 approaches
There are several ways by which these models can be combined to integrate experience-based models and physics-based models, in
build a more efficient degradation model. In general, Liao and which the output of the experience-based model is often used to
Köttig [35] categorized the various hybrid approaches into following enhance the physics-based model. Byington et al. [131] presented
five types: (i) HM1: experience-based model + data-driven model; (ii) a methodology for the RUL prediction of aircraft actuator compo-
HM2: experience-based model + physics-based model; (iii) HM3: nents, where the fuzzy logic was used to quantify the level of
data-driven model of type + data-driven model of another type; (iv) damage (damage index), and Kalman filtering was utilized to

(a) (b)
Fig. 15 Dam performance model using hybrid degradation model (a) X bar and (b) range chart for principal strains

021005-14 / Vol. 4, MAY 2021 Transactions of the ASME

 Table 4 Application of various degradation models for prognosis

Model References Mathematical techniques used Applications

Physics-based
[36] A general framework Remaining useful life
[30] A review Rolling element bearing
[8] Deterioration modeling of structural systems Corrosion and fatigue
[40] Fick’s diffusion law Concrete structures
Data-driven
Bivariate process [91] Adaptive Bayesian decision Replacement policy
[87] Objected-oriented Bayesian networks Time series
[92] Bayesian Network concrete bridges
[93] Kalman filter Conditional residual life
[94] Kalman filter Aircraft power generators
[95,96] Particle filter Fault indicator, fatigue degradation
[97] Stochastic filter Remaining useful life
Cumulative [44–48] Cumulative degradation Ball bearings, composite laminates, fatigue life, fiber-reinforced
damage plastics, and railway wheel damage.
[49] Dependent competing failure Reliability model
[50] Degradation and shock loads Multi-component system reliability model
[51] Dependent competing risk model System subject to shocks
[52] Gradual and shock deterioration processes Life-cycle analysis
Markov [100] Hidden Markov model Diagnosis and prognosis
processes
[138,139] Mixture of Gaussian HMM RUL estimation
[105,140,141] Hidden semi-Markov model Diagnosis & prognosis of pumps
[101,142,143] Hidden Markov model Rolling element bearing
[144–146] Markov model Bridge deck systems
[147] Semi-Markov model Bridge deterioration modeling
Gamma process [64] Theoretical aspects Inspection and maintenance
[67,148,149] Time-dependent reliability Deteriorating structures
[69] Covariates and random effects Fatigue-crack-growth
[20,150,151] Parameters and RUL estimation Corrosion of nuclear pipes
[68] Calculation of confidence limits Creep of concrete
[152] Bivariate degradation Fatigue crack
[153] With random effects Laser degradation
[71,72,154] Condition based maintenance Numerical example
[155] Integration of SHM data Creep deformations of bridges
[148] Temporal variability and uncertainties Concrete sewer pipes
[156] Aging and shock events Bilinear structural systems
Weiner process [62] Time-scale transformation Heating cable
[63] With measurement error Wear of magnetic heads
[60] With random effects bridge beam degradation
[61] RUL using recursive filter Gyros used for navigation
[157] Bivariate Weiner process Aluminum reduction cells
[158] Time correlated structure Resistors and sliding metal
[159] Real-time reliability evaluation Capacitance loss over time
Inverse Gaussian [160] With random-drift Optimal CBM policy
[60] Maximum likelihood estimation Laser devices
[74] Random effects, explanatory variables Laser devices
[161] Bayesian method using inspection data Corrosion growth modeling
[75] A Bayesian perspective GaAs laser degradation data
Other models [117–119] Two-phase degradation Rolling element bearing
[120] Two-phase Wiener and gamma Liquid coupling diode
[121] Multiphase Wiener High storage capacitor
[77–79] Time series model Rotating machinery
[76,80] Long memory models Long-term time series data
[106] Neural network model Rolling contact thrust bearings
[108] Neural network model Highway culverts
[109] Recurrent neural network model and support Pavement performance
vector classifier
[110] Elman neural network model Bridge elements
[107,111– Various machine learning techniques Bridges, water and sewer pipes
116]
Knowledge-based
Expert system [31,122,123] Diagnosis, prognosis, and monitoring Industrial systems
Fuzzy logic [124– Monitoring, diagnosis, and prediction Dam, boiler, bridge high-rise building
127,162]
Hybrid-models
[129] Hybrid approach using Wiener process Cutting tool
[35] Categorization of hybrid approaches Five types
[130] Expert system and neural network Gear box
[132] Time series and statistical process control Multi-storey buildings

Journal of Nondestructive Evaluation, Diagnostics MAY 2021, Vol. 4 / 021005-15

and Prognostics of Engineering Systems
 Table 4 Continued

Model References Mathematical techniques used Applications

[131] Fuzzy logic and Kalman filter Aircraft actuator components

[133] Physics–statistics-based Degradation of accelerometers
[134] Data-driven and particle filter Dynamic system state forecasting
[135] State-space and Gaussian process Fatigue crack growth
[136] Recursive Bayesian approach Fatigue damage prognosis

Table 5 Advantages and drawbacks of various prognosis approaches

Advantages Drawbacks

Physics-based approaches
• Among all the approaches, the physics-based approach provides most • An detailed knowledge of system and underlying degradation process is
accurate and precise RUL estimate required
• For physics-based models, an analytical expression for degradation • These approaches make several assumptions about system to formulated
phenomenon, and a closed form expression for end-life can be derived, the problem, which may not be the case for a unit operating in the field
which can be used for effective maintenance planning
• In this approach, the effect of different variables on degradation process can • These approaches consider degradation as a deterministic process, and
be investigated. Hence, by controlling the some of the variables system’s doesn’t account for uncertainties present due to measurement errors,
life can be increased unit-to-unit variability, and material heterogeneity
• These models are formulated using the fundamental engineering concepts • It is not always possible to formulate the degradation process in terms of
and help to physically understand the underlying degradation mechanism mathematical equations, since the underlying process is complex and
and different modes of system failure unknown
• The physics-based model require relatively less data for parameters • In some cases, it requires finite element modeling, which is
estimation compared to other models such as data-driven model or computationally expensive, if the system is complex
experience- based model
Knowledge-based approaches
Expert system
• Expert system is simple to develop and easy to understand • It depends upon the quality of input knowledge from the domain experts
• Expert system rules can be easily reviewed, and even edited by domain • The model requires a significant number of rules that has to be updated
experts rather than IT experts with time for accurate results
Fuzzy logic
• Fuzzy logic requires relatively fewer rules compared to Expert system • It requires domain knowledge to develop the fuzzy rules
• Input can be imprecise since, the rules are given in term of probability • Assignment of weight to the input variables is difficult decision
• Both ES and FL are easy to maintain. Also, a rapid prototype can be created • It hard to obtain the domain knowledge and subsequently convert them
with a few rules in days rather than the months or years into rules
Data-driven approaches
General comment
• This approach is general and can be used to model any kind of degradation • Many researchers considers this approach as a black-box, which totally
process, once the degradation data are available relies on the data without much knowing about the underlying physics
• With the choice of different functional form, it is possible to generate a • Domain knowledge can be integrated only if the parameters are estimated
wide verity of data-driven degradation models, e.g., random variable, in Bayesian manner
stochastic, Markov and time series
• The various machine learning algorithms such as SVM, NN, clustering, • Most of the data-driven methods require training data collected for all
GMM, and SPC can directly be applied to the data possible operating and loading conditions for an accurate prognosis
Random variable model
• A close form expression for the CDF of failure time, T can be derived • A fixed functional form for a particular degradation process does account
for the variations in environmental and loading conditions
• With the change of functional form a large number of degradation • This model does not account for temporal uncertainty, i.e., uncertainty
processes can modeled associated with the evolution or progression of deterioration over time
Stochastic models:
Markov model
• The mathematics of Markov model is well established and can be used to • It assumes a single monotonic and non-temporal failure degradation
model various the failure scenarios pattern
• Markov model can be applied with the incomplete data sets. Also, the
knowledge of underlying degradation process is not required
• Markov model is computationally efficient, once developed. The calibrated • If time spent in a given state is not exponentially distributed then more
model can be used to predict the RUL along with its confidence interval complex semi-Markov model is required
Hidden Markov model and hidden semi-Markov model
• HMM models the degradation of a system in phases (i.e., hidden states), so • Data corresponding to various states of system are required to train the
the knowledge of continuous degradation path is not necessary HMM model, which increases rapidly with increasing number of hidden
states
• Both HMM and HSMM uses indirect measurements (e.g., vibration, • The suitable computational tools (or software) for HMM and HSMM are
acoustic etc.) obtained from the system not available
• The details of failure progression mechanism is not required • Both are computationally intensive particularly in case if modeling
requires a large number of hidden states
Lifetime data-based approaches
• Lifetime data based approach is most popular among the reliability • This approach does not account for the variation in operating and loading
engineer due to its simplicity compared to the other approaches condition of an infrastructure

021005-16 / Vol. 4, MAY 2021 Transactions of the ASME

 Table 5 Continued

Advantages Drawbacks

• This approach is computationally in-expensive. The calculations can be • Very often, for long-life and high reliable infrastructure lifetime data
performed even on a hand-held calculator remain unavailable
• This method particularly useful in cases where the condition monitoring • In sample where the lifetime data are relatively sparse, reliability estimates
data is difficult to obtain can be overly pessimistic
• This is a population-based approach particularly useful for maintenance • This approach does not focus on the degradation of a particular unit
planning of low cost units
Hazard model
• It integrates the lifetime data and condition monitoring data for reliability • All relevant covariates must be included in the model
assessment
• Softwares are available for computations • Historical lifetime failure data are required
Time series model
• Advanced ARMA model can be applied for non-stationary data • Basic ARMA models assumes that the data are stationary, which is often
not the case
• Lifetime failure data is not required for time series model • Measurement noise present in the signal can result in an inappropriate
model-order selection
• Like other data-driven models time-series approach also does require full • A significant amount of data is required for model parameters estimation
understanding of degradation mechanism
• The various software and computational tools are freely available for fitting • Unlike proportional hazard model, TS model doesnot include the lifetime
a time-series model data and covariate information in analysis
• Time series model can provide accurate short-time forecast • Long-time forecasts using TS model is less reliable
Change point-based model
• Model predicts more accurate RUL when degradation is in the second • Detection of a change point in real-time is a challenging task and can often
phase result in a false alarm
• In this model, prior knowledge can be integrated for improved maintenance • For many long life infrastructure, the prior knowledge about the change
decision of a monitored unit point location is not unavailable
Hybrid approaches
• This approach is useful where limited data is available and some domain • Often two different model one for estimation, and other for prediction is
knowledge partial degradation mechanism is understood required
• A large class of models are available for this approach • In some cases, it may be challenging to integrate the two different types of
model

predict the progression of the damage. In HM3 categories, two differ- to detect any abnormal behavior in the dam responses in real-time. The
ent data-driven model can be fused to come up with a more efficient proposed method is applied to the strain and displacement data of
hybrid model. Here, the first model can be used to estimate while the Tsankov Kamak dam located in Bulgaria. The readers can refer the pre-
second can be used to predict the system state. Recently, Prakash and vious study done by the authors [137] for the details of dam, sensor
Narasimhan [132] used combined time series model in conjunction placement, and data acquisition.
with statistical process control chart for remaining useful life predic- To monitor the health of this dam, several types of instruments
tion of multi-storey buildings. Among all hybrid models, HM4-type were installed across the dam, and data were collected using an
hybrid degradation model is widely applied in the literature. For extensive data collection system. Two of such sensors strain
example, Xu et al. [133] developed a physics–statistics-based degra- gauge and seismic tri-axial accelerometer are shown in Fig. 13.
dation model for reliability predictions and illustrated for data The six strain components (after subtracting the zero-stress strain
obtained from accelerometers for accelerated degradation test. Liu measurements) corresponding to left and right block of crest (top
et al. [134] integrated data-driven prognostics together with a middle block of dam) are shown in Fig. 14, which are used to cal-
physics-based particle filter for system state prediction; Mohanty culate the principal strains and their orientations. In Fig. 14, the sea-
et al. [135] combined a physics-based state space model and a data- sonal trends in strain measurements for both the blocks can clearly
driven kernel-based Gaussian process regression model for fatigue be seen. Moreover, a decreasing trend with time which can be con-
crack growth modeling in metallic alloys. Recently, Paris crack sidered as irreversible is also present for XZ, YZ, and ZZ directions.
growth law in conjunction with the SHM data is utilized to model To model both the seasonality and trend in the data,a hybrid
the fatigue-induced damage in a recursive Bayesian manner by HSTT model is considered as follows:
Gobbato et al. [136]. Finally, all the three approaches, i.e.,
experience-based, data-driven, and physics-based, can be utilized y = yH + ys + yT + yt + ϵ (26)
to built a HM5-type hybrid model. Hybrid approaches in the HM5
type is impractical to implement due to the difficulty that might be where y is the dam response; yH, ys, yT, and yt are the contributions
encountered by each type of model. due to hydrostatic load, seasonal effect, variation in temperature,
and age-related deterioration were applied. The error term ϵ is
assumed to be independent and identically distributed. The follow-
ing functional forms for yH, ys, yt are considered:
5.1 An Application: Hybrid Model. The first author of this
paper has successfully implemented a hybrid model (HM3 type) for


N

M
dam health monitoring. This hybrid prediction model utilizes hydro- yH (H) = ai H i ; yT (T) = bi Tii ;
static, seasonal, temperature and time response data collected from i=1 i=1
the dam and predicts the dam performance during initial service life.
The dam responses such as strain and crest displacement are modeled yt (t) = c1 t + c2 exp(−t) (27)
using recursive least squares while the thresholds are set using univari- ys (θ) = d1 sin(θ) + d2 cos(θ) + d3 sin(θ) cos(θ);
ate and multivariate control charts, based on residuals between the mea-
sured and predicted responses. Furthermore, both the model parameters 2πt
with θ =
and thresholds are updated as more data becomes available, which helps 24 × 365

Journal of Nondestructive Evaluation, Diagnostics MAY 2021, Vol. 4 / 021005-17

and Prognostics of Engineering Systems
where ai, bi, c1, c2, d1, d2, and d3 are constants, H is the reservoir to multiple failure modes and their interdependence are
water level, T is the temperature, t is the time in hours from the required for prognosis of complex systems.
beginning of dam operation, and N, M are the polynomial order Some of the models such as hidden semi-Markov change point-
with respect to H and T, respectively. This model was fitted to based model are yet to be explored for prognosis of critical
the two principal strains (ϵp1, ϵp2,), and residuals were obtained. infrastructure. Only a few researchers have made attempt in
The residuals contain important information for threshold setting this area which is limited to the system where run-to-failure
and anomaly detection. The univariate mean (to detect the mean condition data is available. In future, more works are needed
shift) and range (to detect the variability) control chart for principal for the long life infrastructure where the condition data is
strain residuals are shown in Fig. 15. The thresholds (upper control either unavailable or monitoring is limited to certain periods
limit UCL1 and lower control limit UCL2) are set based on normal- only.
ity assumption of residuals. The mean chart shows that the dam is In future, more emphasis should be given on hybrid degradation
functioning normally. However, there are a few false alarms in models, which takes the advantages of physics-based, data-
the range chart, which is mostly because of limited data set of 44 driven, and knowledge-based approaches.
months, with more data the model parameters, this rate will Finally, very few researches have made attempts toward the inte-
decrease. In this way, a hybrid model can be used to monitor the gration of degradation modeling for maintenance planning
performance of a dams. and decision making. Such models if employed in real
A summary of literature for practical applications of various time will not only reduce the maintenance cost but will
degradation models along with the techniques used are presented also avoid catastrophic failures.
in Table 4, while their advantages and disadvantages are summar-
ized in Table 5.
Acknowledgment
There is no funding source for this review paper.
6 Conclusions, Challenges, and Future Research
In this paper, an attempt has been made to review and summarize Conflict of Interest
the recent degradation modeling developments for prognosis of the
critical infrastructures. Based on the present review, the following There are no conflicts of interest.
conclusions can be made:
All prognosis approaches requires some kind of data, and some References
knowledge about the underlying degradation process. [1] Nelson, W. B., 2009, Accelerated Testing: Statistical Models, Test Plans, and
However, the extent of knowledge and data required Data Analysis, Vol. 344, John Wiley & Sons, Hoboken, NJ.
[2] Escobar, L. A., and Meeker, W. Q., 2006, “A Review of Accelerated Test
depends upon the specific approach. The lifetime-based Models,” Stat. Sci., 21(4), pp. 552–577.
approach requires minimal information among all [3] van Noortwijk, J. M., 2009, “A Survey of the Application of Gamma Processes
approaches followed by other data-driven approaches — in Maintenance,” Reliab. Eng. Syst. Safety, 94(1), pp. 2–21.
which require SHM data for training. The knowledge-based [4] Gorjian, N., Ma, L., Mittinty, M., Yarlagadda, P., and Sun, Y., 2010, “A Review
on Degradation Models in Reliability Analysis,” Engineering Asset Lifecycle
approach such as expert and fuzzy systems requires a Management, J. Mathew, D. Kiritsis, A. Koronios, and C. Emmanouilidis,
medium amount of knowledge and data; while physics-based eds., Springer, London, pp. 369–384.
models require comprehensive knowledge of degradation [5] Si, X. S., Wang, W., Hu, C. H., and Zhou, D. H., 2011, “Remaining Useful Life
mechanisms and environmental factors that influences the Estimation—A Review on the Statistical Data Driven Approaches,”
Eur. J. Operat. Res., 213(1), pp. 1–14.
rate of degradation. Hybrid approaches often requires [6] Ahmad, R., and Kamaruddin, S., 2012, “An Overview of Time-Based and
expert knowledge, lifetime data, prior degradation informa- Condition-Based Maintenance in Industrial Application,” Comput. Indust.
tion,and SHM data. The majority of recent degradation Eng., 63(1), pp. 135–149.
related deep learning studies still focus on defects detection [7] Ye, Z.-S., and Xie, M., 2015, “Stochastic Modelling and Analysis of
Degradation for Highly Reliable Products,” Appl. Stoch. Models Business
and condition assessment. Applications of deep learning Indus., 31(1), pp. 16–32.
methods to damage prognosis are still scarce. [8] Biondini, F., and Frangopol, D. M., 2016, “Life-Cycle Performance of
The accuracy of prognosis results (in terms of RUL confidence Deteriorating Structural Systems Under Uncertainty,” J. Struct. Eng., 142(9),
interval) depends upon the extent of data and knowledge p. F4016001.
[9] Shahraki, A. F., Yadav, O. P., and Liao, H., 2017, “A Review on Degradation
used in the model. Hence, physics-based approaches often Modelling and Its Engineering Applications,” Int. J. Performability Eng.,
give most accurate results, while the lifetime data-based 13(3), p. 299.
approaches yield least accurate ones. [10] Alaswad, S., and Xiang, Y., 2017, “A Review on Condition-Based Maintenance
As different approaches require different amount of data and Optimization Models for Stochastically Deteriorating System,” Reliab. Eng.
Syst. Safety, 157(1), pp. 54–63.
knowledge, and their results present different level of accu- [11] Mevissen, F., and Meo, M., 2019, “A Review of NDT/Structural Health
racy and precision, industrial requirements need to be Monitoring Techniques for Hot Gas Components in Gas Turbines,” Sensors,
clearly established before a proper model is selected. 19(3), p. 711.
From a computational point of view, physics-based models are [12] McCann, D., and Forde, M., 2001, “Review of NDT Methods in the Assessment
of Concrete and Masonry Structures,” NDT E Int., 34(2), pp. 71–84.
most expensive due to mathematical modeling of degrada- [13] Alani, A. M., Aboutalebi, M., and Kilic, G., 2014, “Integrated Health
tion process which often requires a finite element model. Assessment Strategy Using NDT for Reinforced Concrete Bridges,” NDT E
The reliability-based approaches are least expensive, which Int., 61(1), pp. 80–94.
can be performed using a simple Microsoft Excel® work- [14] Drewry, M. A., and Georgiou, G., 2007, “A Review of NDT Techniques for
Wind Turbines,” Insight-Non-Destructive Test. Cond. Monit., 49(3), pp. 137–
sheet. Similarly, in case of knowledge-based approaches, 141.
the calculations are significantly reduced because of [15] Gholizadeh, S., 2016, “A Review of Non-Destructive Testing Methods of
IF-THEN statements, which makes computer coding rela- Composite Materials,” XV Portuguese Conference on Fracture, Paço de
tively straight forward. The computational expense for data- Arcos, Portugal, Feb. 10–12.
[16] Moles, M., Dubé, N., Labbé, S., and Ginzel, E., 2005, “Review of Ultrasonic
driven methods is much less than physics-based model but Phased Arrays for Pressure Vessel and Pipeline Weld Inspections,” ASME
more than knowledge-based model and hybrid model. J. Pressure Vessel Technol., 127(3), pp. 351–356.
Most of the existing literature considers a single failure mode (or [17] Lu, C. J., and Meeker, W. O., 1993, “Using Degradation Measures to Estimate a
at most two) for degradation modeling and subsequent RUL Time-to-Failure Distribution,” Technometrics, 35(2), pp. 161–174.
[18] Pandey, M., Lu, D., and Komljenovic, D., 2009, “The Impact of Probabilistic
prediction. However, for a complex system, often multiple Modelling on Predicting the Remaining Life of Pipes in Nuclear Plants,” 17th
failure modes exist that need to be accounted in the modeling International Conference on Nuclear Engineering, Brussels, Belgium,
for accurate prediction. Hence, more research that is related American Society of Mechanical Engineers, pp. 503–511.

021005-18 / Vol. 4, MAY 2021 Transactions of the ASME

[19] Pandey, M. D., 1998, “Probabilistic Models for Condition Assessment of Oil [49] Rafiee, K., Feng, Q., and Coit, D. W., 2014, “Reliability Modeling for
and Gas Pipelines,” NDT E Int., 31(5), pp. 349–358. Dependent Competing Failure Processes With Changing Degradation Rate,”
[20] Yuan, X.-X., Pandey, M., and Bickel, G., 2008, “A Probabilistic Model of Wall IIE Trans., 46(5), pp. 483–496.
Thinning in Candu Feeders Due to Flow-Accelerated Corrosion,” Nucl. Eng. [50] Song, S., Coit, D. W., Feng, Q., and Peng, H., 2014, “Reliability Analysis for
Des., 238(1), pp. 16–24. Multi-Component Systems Subject to Multiple Dependent Competing Failure
[21] Lin, P., and Yuan, X.-X., 2020, “Development of Performance Measures for Processes,” IEEE Trans. Reliab., 63(1), pp. 331–345.
Pedestrian Sidewalk Asset Management,” Infrastructure Asset Management, [51] Wang, Z., Huang, H.-Z., Li, Y., and Xiao, N.-C., 2011, “An Approach to
7(3), pp. 190–200. Reliability Assessment Under Degradation and Shock Process,” IEEE Trans.
[22] Al-Hussein, A., and Haldar, A., 2017, “Structural Damage Prognosis of Reliab., 60(4), pp. 852–863.
Three-dimensional Large Structural Systems,” Struct. Infrastructure Eng., [52] Iannacone, L., and Gardoni, P., 2019, “Stochastic Differential Equations for the
13(1), pp. 1–13. Deterioration Processes of Engineering Systems,” 13th International Conference
[23] Jin, G., Matthews, D. E., and Zhou, Z., 2013, “A Bayesian Framework for on Applications of Statistics and Probability in Civil Engineering, Seoul, South
On-Line Degradation Assessment and Residual Life Prediction of Secondary Korea, ICASP13, pp. 1663–1670.
Batteries in Spacecraft,” Reliab. Eng. Syst. Safety, 113(1), pp. 7–20. [53] Shah, Y. U., Jain, S., Tiwari, D., and Jain, M., 2013, “Development of Overall
[24] Wang, V. Z., Pease, T., and Robinson, S., 2015, “Statistical Damage Prognosis Pavement Condition Index for Urban Road Network,” Proc.-Soc. Behav. Sci.,
for in-service Civil Structures Against Hazards: Formulations and Applications,” 104(12), pp. 332–341.
J. Eng. Mech., 142(3), p. 04015090. [54] Madanat, S., Mishalani, R., and Ibrahim, W. H. W., 1995, “Estimation of
[25] Li, Z., and Zhang, Y., 2014, “Extreme Value Theory-based Structural Health Infrastructure Transition Probabilities From Condition Rating Data,”
Prognosis Method Using Reduced Sensor Data,” Struct. Infrastruct. Eng., J. Infrastruct. Syst., 1(2), pp. 120–125.
10(8), pp. 988–997. [55] Tsuda, Y., Kaito, K., Aoki, K., and Kobayashi, K., 2006, “Estimating
[26] Jardine, A. K. S., Lin, D., and Banjevic, D., 2006, “A Review on Machinery Markovian Transition Probabilities for Bridge Deterioration Forecasting,”
Diagnostics and Prognostics Implementing Condition-Based Maintenance,” Struct. Eng./Earthquake Eng., 23(2), pp. 241s–256s.
Mech. Syst. Signal Proc., 20(7), pp. 1483–1510. [56] Kharoufeh, J. P., Solo, C. J., and Ulukus, M. Y., 2010, “Semi-Markov Models
[27] Sikorska, J., Hodkiewicz, M., and Ma, L., 2011, “Prognostic Modelling Options for Degradation-Based Reliability,” IIE Trans., 42(8), pp. 599–612.
for Remaining Useful Life Estimation by Industry,” Mech. Syst. Signal Proc., [57] Compare, M., Martini, F., Mattafirri, S., Carlevaro, F., and Zio, E., 2016,
25(5), pp. 1803–1836. “Semi-Markov Model for the Oxidation Degradation Mechanism in Gas
[28] Das, S., and Saha, P., 2018, “Structural Health Monitoring Techniques Turbine Nozzles,” IEEE Trans. Reliab., 65(2), pp. 574–581.
Implemented on IASC-ASCE Benchmark Problem: A Review,” J. Civil [58] Karlin, S., and Taylor, H. M., 1975, A First Course in Stochastic Processes,
Struct. Health Monit., 8(4), pp. 689–718. Academic, San Diego.
[29] Lei, Y., Li, N., Guo, L., Li, N., Yan, T., and Lin, J., 2018, “Machinery Health [59] Doksum, K. A., and Hbyland, A., 1992, “Models for Variable-Stress
Prognostics: A Systematic Review From Data Acquisition to RUL Accelerated Life Testing Experiments Based on Wener Processes and the
Prediction,” Mech. Syst. Signal Proc., 104(5), pp. 799–834. Inverse Gaussian Distribution,” Technometrics, 34(1), pp. 74–82.
[30] Cubillo, A., Perinpanayagam, S., and Esperon-Miguez, M., 2016, “A Review of [60] Wang, X., 2010, “Wiener Processes With Random Effects for Degradation
Physics-Based Models in Prognostics: Application to Gears and Bearings of Data,” J. Multi. Anal., 101(2), pp. 340–351.
Rotating Machinery,” Adv. Mech. Eng., 8(8), p. 1687814016664660. [61] Si, X.-S., Wang, W., Hu, C.-H., Chen, M.-Y., and Zhou, D.-H., 2013, “A
[31] Peng, Y., Dong, M., and Zuo, M. J., 2010, “Current Status of Machine Wiener-Process-Based Degradation Model With a Recursive Filter Algorithm for
Prognostics in Condition-Based Maintenance: A Review,” Int. J. Adv. Manuf. Remaining Useful Life Estimation,” Mech. Syst. Signal Proc., 35(1), pp. 219–237.
Technol., 50(1-4), pp. 297–313. [62] Whitmore, G., and Schenkelberg, F., 1997, “Modelling Accelerated Degradation
[32] Zhang, Z., Si, X., Hu, C., and Lei, Y., 2018, “Degradation Data Analysis and Data Using Wiener Diffusion With a Time Scale Transformation,” Life. Data
Remaining Useful Life Estimation: A Review on Wiener-Process-Based Anal., 3(1), pp. 27–45.
Methods,” Eur. J. Operat. Res., 271(3), pp. 775–796. [63] Ye, Z.-S., Wang, Y., Tsui, K.-L., and Pecht, M., 2013, “Degradation Data
[33] Si, X.-S., Wang, W., Hu, C.-H., and Zhou, D.-H., 2011, “Remaining Useful Life Analysis Using Wiener Processes With Measurement Errors,” Reliab., IEEE
Estimation—A Review on the Statistical Data Driven Approaches,” Trans. on, 62(4), pp. 772–780.
Eur. J. Operat. Res., 213(1), pp. 1–14. [64] Van Noortwijk, J., 2009, “A Survey of the Application of Gamma Processes in
[34] Heng, A., Zhang, S., Tan, A. C., and Mathew, J., 2009, “Rotating Machinery Maintenance,” Reliab. Eng. Syst. Safety, 94(1), pp. 2–21.
Prognostics: State of the Art, Challenges and Opportunities,” Mech. Syst. [65] Guo, B., and Tan, L., 2009, “Reliability Assessment of Gamma Deteriorating
Signal Proc., 23(3), pp. 724–739. System Based on Bayesian Updating,” 8th International Conference on
[35] Liao, L., and Köttig, F., 2014, “Review of Hybrid Prognostics Approaches for Reliability, Maintainability and Safety, Chengdu, China, July 20–24, IEEE,
Remaining Useful Life Prediction of Engineered Systems, and An Application pp. 429–432.
to Battery Life Prediction,” IEEE Trans. Reliab., 63(1), pp. 191–207. [66] Ye, Z.-S., Xie, M., Tang, L.-C., and Chen, N., 2014, “Semiparametric
[36] Luo, J., Bixby, A., Pattipati, K., Qiao, L., Kawamoto, M., and Chigusa, S., 2003, Estimation of Gamma Processes for Deteriorating Products,” Technometrics,
“An Interacting Multiple Model Approach to Model-Based Prognostics,” IEEE 56(4), pp. 504–513.
International Conference on Systems, Man and Cybernetics, Washington, DC, [67] Edirisinghe, R., Setunge, S., and Zhang, G., 2013, “Application of Gamma
Vol. 1, IEEE, pp. 189–194. Process for Building Deterioration Prediction,” J. Perform. Const. Facilities,
[37] Tripura, T., Gogoi, A., and Hazra, B., 2020, “An ITO-Taylor Weak 3.0 Method 27(6), pp. 763–773.
for Stochastic Dynamics of Nonlinear Systems,” Appl. Math. Model., 86(10), [68] Cinlar, E., Bazant, Z. P., and Osman, E. M., 1977, “Stochastic Process for
pp. 115–141. Extrapolating Concrete Creep,” J. Eng. Mech., 103(6), pp. 1069–1088.
[38] Tripura, T., Bhowmik, B., Pakrashi, V., and Hazra, B., 2020, “Real-Time [69] Lawless, J., and Crowder, M., 2004, “Covariates and Random Effects in a
Damage Detection of Degrading Systems,” Struct. Health. Monit., 19(3), Gamma Process Model With Application to Degradation and Failure,”
pp. 810–837. Lifetime Data Anal., 10(3), pp. 213–227.
[39] Paris, P., and Erdogan, F., 1963, “A Critical Analysis of Crack Propagation [70] Van Noortwijk, J., Kallen, M., and Pandey, M., 2005, “Gamma Processes for
Laws,” ASME J. Basic. Eng., 85(4), pp. 528–533. Time-Dependent Reliability of Structures,” Adv. Safety Reliab., Proc. ESREL,
[40] Biondini, F., Bontempi, F., Frangopol, D. M., and Malerba, P. G., 2006, 12, pp. 1457–1464.
“Probabilistic Service Life Assessment and Maintenance Planning of Concrete [71] Grall, A., Bérenguer, C., and Dieulle, L., 2002, “A Condition-Based
Structures,” J. Struct. Eng., 132(5), pp. 810–825. Maintenance Policy for Stochastically Deteriorating Systems,” Reliab. Eng.
[41] Kim, N.-H., An, D., and Choi, J.-H., 2016, Prognostics and Health Management Syst. Safety, 76(2), pp. 167–180.
of Engineering Systems: An Introduction, Springer, Gewerbestrasse 11, 6330 [72] Liao, H., Elsayed, E. A., and Chan, L.-Y., 2006, “Maintenance of Continuously
Cham, Switzerland. Monitored Degrading Systems,” Eur. J. Opera. Res., 175(2), pp. 821–835.
[42] Nikulin, M., Limnios, N., Balakrishnan, N., Kahle, W., and Huber-Carol, C., [73] Wang, X., and Xu, D., 2010, “An Inverse Gaussian Process Model for
2010, “Advances in Degradation Modeling: Applications to Reliability,” Degradation Data,” Technometrics, 52(2), pp. 188–197.
Survival Analysis, and Finance, Springer/Birkhauser, Boston, MA, p. 416. [74] Ye, Z.-S., and Chen, N., 2014, “The Inverse Gaussian Process As a Degradation
[43] Yuan, X.-X., 2007, “Stochastic Modeling of Deterioration in Nuclear Power Model,” Technometrics, 56(3), pp. 302–311.
Plant Components,” Ph.D. thesis, University of Waterloo, Waterloo, ON, [75] Peng, W., Li, Y.-F., Yang, Y.-J., Huang, H.-Z., and Zuo, M. J., 2014, “Inverse
Canada. Gaussian Process Models for Degradation Analysis: A Bayesian Perspective,”
[44] Zaretsky, E. V., “A. Palmgren Revisited: A basis for Bearing Life Prediction,” Reliab. Eng. Syst. Safety, 130(10), pp. 175–189.
NASA Technical Memorandum 107440. [76] Box, G. E., Jenkins, G. M., Reinsel, G. C., and Ljung, G. M., 2015, Time Series
[45] Subramanian, S., Reifsnider, K., and Stinchcomb, W., 1995, “A Cumulative Analysis: Forecasting and Control, John Wiley & Sons, Hoboken, NJ.
Damage Model to Predict the Fatigue Life of Composite Laminates [77] Wu, W., Hu, J., and Zhang, J., 2007, “Prognostics of Machine Health Condition
Including the Effect of a Fibre-Matrix Interphase,” Int. J. Fatigue., 17(5), Using An Improved Arima-Based Prediction Method,” 2nd IEEE Conference on
pp. 343–351. Industrial Electronics and Applications, Harbin, China, May 23–25, IEEE,
[46] Hwang, W., and Han, K., 1986, “Cumulative Damage Models and Multi-Stress pp. 1062–1067.
Fatigue Life Prediction,” J. Compos. Mater., 20(2), pp. 125–153. [78] Yan, J., Koc, M., and Lee, J., 2004, “A Prognostic Algorithm for Machine
[47] Yao, W., and Himmel, N., 2000, “A New Cumulative Fatigue Damage Performance Assessment and Its Application,” Prod. Planning Control, 15(8),
Model for Fibre-Reinforced Plastics,” Compos. Sci. Technol., 60(1), pp. 59–64. pp. 796–801.
[48] Ekberg, A., Bjarnehed, H., and Lundbéan, R., 1995, “A Fatigue Life Model for [79] Kobayashi, K., Kaito, K., and Kazumi, K., 2014, “Deterioration Forecasting of
General Rolling Contact With Application to Wheel/Rail Damage,” Fatigue. Joint Members Based on Longterm Monitoring Data,” EURO J. Trans. Logist.,
Fract. Eng. Mater. Struct., 18(10), pp. 1189–1199. 4(1), pp. 5–30.

Journal of Nondestructive Evaluation, Diagnostics MAY 2021, Vol. 4 / 021005-19

and Prognostics of Engineering Systems
 [80] Kobayashi, K., Kaito, K., and Mizutani, D., 2017, “Deterioration Prediction of [109] Tabatabaee, N., Ziyadi, M., and Shafahi, Y., 2013, “Two-stage Support Vector
Infrastructures with Time Series Data Considering Long Memory Effect,” Classifier and Recurrent Neural Network Predictor for Pavement Performance
IABSE International Association for Bridge and Structural Engineering, Modeling,” J. Infrastruct. Syst., 19(3), pp. 266–274.
Zurich, Switzerland, Sept., 2015. [110] Lee, J., Guan, H., Loo, Y.-C., and Blumenstein, M., 2014, “Development of a
[81] Liao, H., and Elsayed, M., 2005, “Optimization of System Reliability Long-Term Bridge Element Performance Model Using Elman Neural
Robustness Using Accelerated Degradation Testing,” 2005, Proceedings of Networks,” J. Infrastruct. Syst., 20(3), p. 04014013.
Annual Reliability and Maintainability Symposium, Alexandria, VA, Jan. 24– [111] Alipour, M., and Harris, D. K., 2020, “A Big Data Analytics Strategy for
27, IEEE, pp. 48–54. Scalable Urban Infrastructure Condition Assessment Using Semi-Supervised
[82] Lydersen, S., and Rausand, M., 1987, “A Systematic Approach to Acclerated Multi-Transform Self-Training,” J. Civil Struct. Health Monitoring, 10(2),
Life Testing,” Reliab. Eng., 18(4), pp. 285–293. pp. 313–332.
[83] Park, C., and Padgett, W., 2005, “Accelerated Degradation Models for Failure [112] Cheng, M. Y., and Hoang, N. D., 2019, “Risk Score Inference for Bridge
Based on Geometric Brownian Motion and Gamma Processes,” Lifetime Data Maintenance Project Using Evolutionary Fuzzy Least Squares Support Vector
Anal., 11(4), pp. 511–527. Machine,” J. Comput. Civil Eng., 28(3), p. 04014003.
[84] Trevisanello, L., Meneghini, M., Mura, G., Vanzi, M., Pavesi, M., Meneghesso, [113] Fiorillo, G., and Nassif, H., 2020, “Improving the Conversion Accuracy
G., and Zanoni, E., 2008, “Accelerated Life Test of High Brightness Light Between Bridge Element Conditions and NBI Ratings Using
Emitting Diodes,” IEEE Trans. Device Mater. Reliab., 8(2), pp. 304–311. Deep Convolutional Neural Networks,” Struct. Infrastruct. Eng., 16(12), pp.
[85] Wang, W., and Dragomir-Daescu, D., 2002, “Reliability Quantification of 1669–1682.
Induction Motors-accelerated Degradation Testing Approach,” Proceedings of [114] Li, P., Jia, X. D., Feng, J. S., Zhu, F., Miller, M., Chen, L. Y., and Lee, J., 2020,
Annual Reliability and Maintainability Symposium, Seattle, WA, IEEE, “A Novel Scalable Method for Machine Degradation Assessment Using Deep
pp. 325–331. Convolutional Neural Network,” Measurement, 151(2), p. 107106.
[86] Zhang, C., Le, M., Seth, B., and Liang, S., 2002, “Bearing Life Prognosis Under [115] Liu, H., and Zhang, Y. F., 2020, “Bridge Condition Rating Data Modeling Using
Environmental Effects Based on Accelerated Life Testing,” Proc. Inst. Mech. Deep Learning Algorithm,” Struct. Infrastruct. Eng., 16(10), pp. 1447–1460.
Eng., Part C: J. Mech. Eng. Sci., 216(5), pp. 509–516. [116] Malek Mohammadi, M., Najafi, M., Kaushal, V., Serajiantehrani, R.,
[87] Weidl, G., Madsen, A. L., and Israelson, S., 2005, “Applications of Salehabadi, N., and Ashoori, T., 2019, “Sewer Pipes Condition Prediction
Object-Oriented Bayesian Networks for Condition Monitoring, Root Cause Models: A State-of-the-Art Review,” Infrastructures, 4(4), p. 64.
Analysis and Decision Support on Operation of Complex Continuous [117] Gebraeel, N., 2006, “Sensory-updated Residual Life Distributions for
Processes,” Comput. Chem. Eng., 29(9), pp. 1996–2009. Components With Exponential Degradation Patterns,” Aut. Sci. Eng., IEEE
[88] Maguluri, G., and Zhang, C.-H., 1994, “Estimation in the Mean Residual Trans., 3(4), pp. 382–393.
Life Regression Model,” J. R. Stat. Soc.: Ser. B (Methodological), 56(3), [118] Chen, N., and Tsui, K. L., 2013, “Condition Monitoring and Remaining Useful
pp. 477–489. Life Prediction Using Degradation Signals: Revisited,” IIE Trans., 45(9),
[89] Yuen, K., Zhu, L., and Tang, N., 2003, “On the Mean Residual Life Regression pp. 939–952.
Model,” J. Stat. Plann. Inference, 113(2), pp. 685–698. [119] Prakash, G., Narasimhan, S., and Pandey, M. D., 2018, “A Probabilistic
[90] Dey, S., and Stori, J., 2005, “A Bayesian Network Approach to Root Approach to Remaining Useful Life Prediction of Rolling Element Bearings,”
Cause Diagnosis of Process Variations,” Int. J. Mach. Tools. Manuf., 45(1), Struct. Health. Monit., 3, p. 1475921718758517.
pp. 75–91. [120] Wang, X., Jiang, P., Guo, B., and Cheng, Z., 2014, “Real-Time Reliability
[91] Kallen, M.-J., and van Noortwijk, J. M., 2005, “Optimal Maintenance Decisions Evaluation for An Individual Product Based on Change-Point Gamma and
Under Imperfect Inspection,” Reliab. Eng. Syst. Safety, 90(2–3), pp. 177–185. Wiener Process,” Q. Reliab. Eng. Int., 30(4), pp. 513–525.
[92] Yuan, M., Liu, Y., Yan, D., and Liu, Y., 2019, “Probabilistic Fatigue Life [121] Feng, J., Sun, Q., and Jin, T., 2012, “Storage Life Prediction for a
Prediction for Concrete Bridges Using Bayesian Inference,” Adv. Struct. Eng., High-Performance Capacitor Using Multi-Phase Wiener Degradation Model,”
22(3), pp. 765–778. Commun. Stat.- Simulation Comput., 41(8), pp. 1317–1335.
[93] Wang, W., 1997, “Subjective Estimation of the Delay Time Distribution in [122] Jackson, P., 1998, Introduction to Expert Systems, Addison-Wesley Longman
Maintenance Modelling,” Eur. J. Operat. Res., 99(3), pp. 516–529. Publishing Co Inc., Boston, MA.
[94] Batzel, T. D., and Swanson, D. C., 2009, “Prognostic Health Management of [123] Liao, S.-H., 2005, “Expert System Methodologies and Applications–a Decade
Aircraft Power Generators,” IEEE Trans. Aeros. Electronic Syst., 45(2), Review From 1995 to 2004,” Expert syst. Appl., 28(1), pp. 93–103.
pp. 473–482. [124] Salvaneschi, P., Cedei, M., and Lazzari, M., 1996, “Applying AI to Structural
[95] Orchard, M. E., and Vachtsevanos, G. J., 2009, “A Particle-Filtering Approach Safety Monitoring and Evaluation,” IEEE Expert, 11(4), pp. 24–34.
for On-Line Fault Diagnosis and Failure Prognosis,” Trans. Inst. Meas. Control, [125] Majidian, A., and Saidi, M., 2007, “Comparison of Fuzzy Logic and Neural
31(3–4), pp. 221–246. Network in Life Prediction of Boiler Tubes,” Int. J. Fatigue., 29(3), pp. 489–498.
[96] Cadini, F., Zio, E., and Avram, D., 2009, “Model-Based Monte Carlo State [126] Symans, M. D., and Kelly, S. W., 1999, “Fuzzy Logic Control of Bridge
Estimation for Condition-Based Component Replacement,” Reliab. Eng. Syst. Structures Using Intelligent Semi-Active Seismic Isolation Systems,”
Safety, 94(3), pp. 752–758. Earthquake Eng. Struct. Dyn., 28(1), pp. 37–60.
[97] Tang, L., DeCastro, J., Kacprzynski, G., Goebel, K., and Vachtsevanos, G., [127] Pourzeynali, S., Lavasani, H., and Modarayi, A., 2007, “Active Control of High
2010, “Filtering and Prediction Techniques for Model-Based Prognosis and Rise Building Structures Using Fuzzy Logic and Genetic Algorithms,” Eng.
Uncertainty Management,” 2010 Prognostics and System Health Management Struct., 29(3), pp. 346–357.
Conference, Macao, China, Jan. 12–14. [128] Ross, T. J., 2009, Fuzzy Logic with Engineering Applications, John Wiley &
[98] Rabiner, L. R., and Juang, B.-H., 1993, Fundamentals of Speech Recognition, Sons, Chichester, West Sussex, UK.
Vol. 14, PTR Prentice Hall , Englewood Cliffs, Hoboken, NJ. [129] Sun, H., Cao, D., Zhao, Z., and Kang, X., 2018, “A Hybrid Approach to Cutting
[99] Boutros, T., and Liang, M., 2011, “Detection and Diagnosis of Bearing and Tool Remaining Useful Life Prediction Based on the Wiener Process,” IEEE
Cutting Tool Faults Using Hidden Markov Models,” Mech. Syst. Signal Trans. Reliab., 67(3), pp. 1294–1303.
Process., 25(6), pp. 2102–2124. [130] Garga, A. K., McClintic, K. T., Campbell, R. L., Yang, C.-C., Lebold, M. S.,
[100] Baruah, P., and Chinnam, R. B., 2005, “HMMs for Diagnostics and Prognostics Hay, T. A., and Byington, C. S., 2001, “Hybrid Reasoning for Prognostic
in Machining Processes,” Int. J. Product. Res., 43(6), pp. 1275–1293. Learning in Cbm Systems,” 2001 IEEE Aerospace Conference Proceedings
[101] Zhang, X., Xu, R., Kwan, C., Liang, S. Y., Xie, Q., and Haynes, L., 2005, “An (Cat. No. 01TH8542), Big Sky, MT, Mar. 10–17.
Integrated Approach to Bearing Fault Diagnostics and Prognostics,” 2005 [131] Byington, C. S., Watson, M., and Edwards, D., 2004, “Data-driven Neural
American Control Conference ThB15., Portland, OR, June 8–10, IEEE, Network Methodology to Remaining Life Predictions for Aircraft Actuator
pp. 2750–2755. Components,” 2004 IEEE Aerospace Conference Proceedings (IEEE Cat. No.
[102] Kobayashi, K., Kaito, K., and Lethanh, N., 2012, “A Statistical Deterioration 04TH8720), Big Sky, MT, Mar. 6–13.
Forecasting Method Using Hidden Markov Model for Infrastructure [132] Prakash, G., and Narasimhan, S., 2017, “Bayesian Two-Phase Gamma Process
Management,” Trans. Res. Part B: Methodol., 46(4), pp. 544–561. Model for Damage Detection and Prognosis,” J. Eng. Mech., 144(2),
[103] Lethanh, N., Kaito, K., and Kobayashi, K., 2014, “Infrastructure Deterioration p. 04017158.
Prediction With a Poisson Hidden Markov Model on Time Series Data,” [133] Xu, D., Wei, Q., Chen, Y., and Kang, R., 2015, “Reliability Prediction Using
J. Infrastruct. Syst., 21(3), p. 04014051. Physics–Statistics-Based Degradation Model,” IEEE Trans. Components,
[104] Dong, M., and He, D., 2007, “A Segmental Hidden Semi-Markov Model Packaging Manufacturing Technol., 5(11), pp. 1573–1581.
(HSMM)-Based Diagnostics and Prognostics Framework and Methodology,” [134] Liu, J., Wang, W., Ma, F., Yang, Y., and Yang, C., 2012, “A Data-Model-Fusion
Mech. Syst. Signal Proc., 21(5), pp. 2248–2266. Prognostic Framework for Dynamic System State Forecasting,” Eng. Appl.
[105] Peng, Y., and Dong, M., 2011, “A Prognosis Method Using Age-Dependent Artificial Intellig., 25(4), pp. 814–823.
Hidden Semi-Markov Model for Equipment Health Prediction,” Mech. Syst. [135] Mohanty, S., Teale, R., Chattopadhyay, A., Peralta, P., and Willhauck, C., 2007,
Signal Proc., 25(1), pp. 237–252. “Mixed Gaussian Process and State-space Approach for Fatigue Crack Growth
[106] Gebraeel, N. Z., and Lawley, M. A., 2008, “A Neural Network Degradation Prediction,” International Workshop on Structural Heath Monitoring, Stanford,
Model for Computing and Updating Residual Life Distributions,” IEEE Trans. CA, Sept. 11–13.
Auto. Sci. Eng., 5(1), pp. 154–163. [136] Gobbato, M., Kosmatka, J. B., and Conte, J. P., 2014, “A Recursive
[107] Tran, H. D., Perera, B., and Ng, A., 2010, “Markov and Neural Network Models Bayesian Approach for Fatigue Damage Prognosis: An Experimental
for Prediction of Structural Deterioration of Storm-Water Pipe Assets,” Validation At the Reliability Component Level,” Mech. Syst. Signal Process.,
J. Infrastruct. Syst., 16(2), pp. 167–171. 45(2), pp. 448–467.
[108] Tatari, O., Sargand, S. M., Masada, T., and Tarawneh, B., 2013, “Neural [137] Prakash, G., Sadhu, A., Narasimhan, S., and Brehe, J.-M., 2018, “Initial Service
Network Approach to Condition Assessment of Highway Culverts: Case Life Data Towards Structural Health Monitoring of a Concrete Arch Dam,”
Study in Ohio,” J. Infrastruct. Syst., 19(4), pp. 409–414. Struct. Control Health Monitoring, 25(1), p. e2036.

021005-20 / Vol. 4, MAY 2021 Transactions of the ASME

[138] Tobon-Mejia, D., Medjaher, K., Zerhouni, N., and Tripot, G., 2011, “Hidden Degradation Measurements,” Proc. Inst. Mech. Eng., Part O: J. Risk Reliab.,
Markov Models for Failure Diagnostic and Prognostic,” Prognostics and 227(4), pp. 425–433.
System Health Management Conference (PHM-Shenzhen), Shenzhen, China, [151] Van Noortwijk, J., and Pandey, M. D., 2003, “A Stochastic Deterioration
May 24–25. Process for Time-dependent Reliability Analysis,” Proceedings of the
[139] Tobon-Mejia, D. A., Medjaher, K., Zerhouni, N., and Tripot, G., 2012, “A Eleventh IFIP WG 7.5 Working Conference on Reliability and Optimization
Data-Driven Failure Prognostics Method Based on Mixture of Gaussian of Structural Systems, Banff, Canada, Nov. 2–5, Taylor & Francis Group.
Hidden Markov Models,” IEEE Trans. Reliab., 61(2), pp. 491–503. [152] Pan, Z., and Balakrishnan, N., 2011, “Reliability Modeling of Degradation of
[140] Dong, M., and He, D., 2007, “Hidden semi-Markov Model-Based Methodology Products With Multiple Performance Characteristics Based on Gamma
for Multi-sensor Equipment Health Diagnosis and Prognosis,” Eur. J. Operat. Processes,” Reliab. Eng. Syst. Safety, 96(8), pp. 949–957.
Res., 178(3), pp. 858–878. [153] Tsai, C. C., Tseng, S. T., and Balakrishnan, N., 2012, “Optimal Design for
[141] Dong, M., He, D., Banerjee, P., and Keller, J., 2006, “Equipment Health Degradation Tests Based on Gamma Processes With Random Effects,” IEEE
Diagnosis and Prognosis Using Hidden semi-Markov Models,” Int. J. Adv. Trans. Reliab., 61(2), pp. 604–613.
Manuf. Technol., 30(7–8), pp. 738–749. [154] Cheng, T., Pandey, M. D., and van der Weide, J. A., 2012, “The Probability
[142] Ocak, H., and Loparo, K. A., 2005, “HMM-Based Fault Detection and Distribution of Maintenance Cost of a System Affected by the Gamma
Diagnosis Scheme for Rolling Element Bearings,” ASME J. Vib. Acoust., Process of Degradation: Finite Time Solution,” Reliab. Eng. Syst. Safety,
127(4), pp. 299–306. 108(12), pp. 65–76.
[143] Purushotham, V., Narayanan, S., and Prasad, S. A., 2005, “Multi-fault Diagnosis [155] Strauss, A., Wan-Wendner, R., Vidovic, A., Zambon, I., Yu, Q., Frangopol,
of Rolling Bearing Elements Using Wavelet Analysis and Hidden Markov D. M., and Bergmeister, K., 2017, “Gamma Prediction Models for Long-Term
Model Based Fault Recognition,” Ndt & E Int., 38(8), pp. 654–664. Creep Deformations of Prestressed Concrete Bridges,” J. Civ. Eng. Manag.,
[144] Morcous, G., 2006, “Performance Prediction of Bridge Deck Systems Using 23(6), pp. 681–698.
Markov Chains,” J. Perfor. Construct. Facilities, 20(2), pp. 146–155. [156] Iervolino, I., Giorgio, M., and Chioccarelli, E., 2013, “Gamma Degradation
[145] Morcous, G., Lounis, Z., and Mirza, M., 2003, “Identification of Environmental Models for Earthquake-Resistant Structures,” Struct. Safety, 45(11), pp. 48–58.
Categories for Markovian Deterioration Models of Bridge Decks,” J. Bridge [157] Whitmore, G. a., Crowder, M. J., and Lawless, J. F., 1998, “Failure Inference
Eng., 8(6), pp. 353–361. From a Marker Process Based on a Bivariate Wiener Model.,” Lifetime Data
[146] Kallen, M. J., and Noortwijk, J. M. V., 2006, “Statistical Inference for Markov Anal., 4(3), pp. 229–251.
Deterioration Models of Bridge Conditions in the Netherlands,” Third [158] Li, J., Wang, Z., Zhang, Y., Fu, H., Liu, C., and Krishnaswamy, S., 2017,
International Conference on Bridge Maintenance, Safety and Management “Degradation Data Analysis Based on a Generalized Wiener Process Subject
(IABMAS) (2006), Porto, Portugal, July 16–19. to Measurement Error,” Mech. Syst. Signal Proc., 94(9), pp. 57–72.
[147] Ng, S. K., and Moses, F., 1998, “Bridge Deterioration Modeling Using [159] Wang, X., Jiang, P., Guo, B., and Cheng, Z., 2014, “Real-time Reliability
Semi-Markov Theory,” Proceedings of ICOSSAR ’97, the 7th International Evaluation With a General Wiener Process-Based Degradation Model,”
Conference on Structural Safety and Reliability, Kyoto, Japan, Nov. 24–28, Q. Reliab. Eng. Int., 30(2), pp. 205–220.
1997, CRC Press, Vol. 1, pp. 113–120. [160] Chen, N., Ye, Z.-S., Xiang, Y., and Zhang, L., 2015, “Condition-Based
[148] Huang, X., and Chen, J., 2014, “Time-Dependent Reliability Model of Maintenance Using the Inverse Gaussian Degradation Model,” Eur. J. Oper.
Deteriorating Structures Based on Stochastic Processes and Bayesian Res., 243(1), pp. 190–199.
Inference Methods,” J. Eng. Mech., 141(3), p. 04014123. [161] Qin, H., Zhang, S., and Zhou, W., 2013, “Inverse Gaussian Process-Based
[149] Pandey, M. D., Yuan, X. X., and van Noortwijk, J. M., 2005, “Gamma Process Corrosion Growth Modeling and Its Application in the Reliability Analysis
Model for Reliability Analysis and Replacement of Aging Structural for Energy Pipelines,” Frontiers Struct. Civil Eng., 7(3), pp. 276–287.
Components,” Ninth International Conference on Structural Safety and [162] Pragalath, H., Seshathiri, S., Rathod, H., Esakki, B., and Gupta, R.,
Reliability, Rome, Italy, June 19–23. 2018, “Deterioration Assessment of Infrastructure Using Fuzzy Logic and
[150] Lu, D., Pandey, M. D., and Xie, W.-C., 2013, “An Efficient Method for the Image Processing Algorithm,” J. Perform. Construct. Facilities, 32(2),
Estimation of Parameters of Stochastic Gamma Process From Noisy p. 04018009.

Journal of Nondestructive Evaluation, Diagnostics MAY 2021, Vol. 4 / 021005-21

and Prognostics of Engineering Systems