Structural and Multidisciplinary Optimization

https://doi.org/10.1007/s00158-021-03026-7

REVIEW PAPER

Modeling, analysis, and optimization under uncertainties: a review

Erdem Acar1   · Gamze Bayrak1 · Yongsu Jung2 · Ikjin Lee2 · Palaniappan Ramu3   · Suja Shree Ravichandran3

Received: 29 December 2020 / Revised: 16 June 2021 / Accepted: 23 July 2021

© The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature 2021

Abstract
Design optimization of structural and multidisciplinary systems under uncertainty has been an active area of research due
to its evident advantages over deterministic design optimization. In deterministic design optimization, the uncertainties of
a structural or multidisciplinary system are taken into account by using safety factors specified in the regulations or design
codes. This uncertainty treatment is a subjective and indirect way of dealing with uncertainty. On the other hand, design
under uncertainty approaches provide an objective and direct way of dealing with uncertainty. This paper provides a review
of the uncertainty treatment practices in design optimization of structural and multidisciplinary systems under uncertainties.
To this end, the activities in uncertainty modeling are first reviewed, where theories and methods on uncertainty categori-
zation (or classification), uncertainty handling (or management), and uncertainty characterization are discussed. Second,
the tools and techniques developed and used for uncertainty modeling and propagation are discussed under the broad two
classes of probabilistic and non-probabilistic approaches. Third, various design optimization methods under uncertainty
which incorporate all the techniques covered in uncertainty modeling and analysis are reviewed. In addition to these in-depth
reviews on uncertainty modeling, uncertainty analysis, and design optimization under uncertainty, some real-life engineering
applications and benchmark test examples are provided in this paper so that readers can develop an appreciation on where
and how the discussed techniques can be applied and how to compare them. Finally, concluding remarks are provided, and
areas for future research are suggested.

Keywords  Uncertainties · Optimization · Modeling · Characterization · Propagation · Analysis · Quantification

1 Introduction of uncertainty modeling, uncertainty analysis, and design

optimization under uncertainty in a single review paper.
1.1 Motivation of the review Hence, this paper aims to (1) provide an in-depth review
of uncertainty treatment practices such as uncertainty mod-
Though uncertainty has been incorporated into many studies eling, uncertainty analysis, and design under uncertainty; (2)
on design and optimization of structural and multidiscipli- suggest areas for future research; and (3) complement exist-
nary systems, the current literature lacks a combined review ing reviews on similar topics such as reliability analysis by
Rackwitz (2001), sampling-based methods for uncertainty
and sensitivity analysis by Helton et al. (2006), reliability-
Responsible Editor: Pingfeng Wang.
based optimization by Valdebenito and Schuëller (2010),
Special Issue dedicated to Dr. Raphael T. Haftka uncertainty handling theories by Li et al. (2012), uncertainty
representation by Zio and Pedroni (2013), and others. In
* Palaniappan Ramu addition to the in-depth review on uncertainty treatment in
[email protected] engineering fields, practical engineering applications and
1
Department of Mechanical Engineering, TOBB University
benchmark test examples are listed in this paper so that read-
of Economics and Technology, 06560 Ankara, Turkey ers can easily understand where to apply the topics explained
2
Department of Mechanical Engineering, Korea Advanced
in the paper or how to compare them.
Institute of Science and Technology, 291 Daehak‑ro,
Yuseong‑gu, Daejeon 34141, Republic of Korea
3
Department of Engineering Design, Indian Institute
of Technology Madras, Chennai 600036, India

13
Vol.:(0123456789)
 E. Acar et al.

1.2 Definition of uncertainty characterization activities are reviewed in this section.

Characterized uncertain inputs are to be propagated
The term uncertainty has different meanings depending on through the analysis models to obtain uncertain outputs.
the domain of application. In economics, Knight (1921) Section 4 provides a review of analysis models, methods,
used the term uncertainty to describe the unquantifiable and tools. Probabilistic techniques (e.g., sampling-based,
knowledge about some possible occurrence, as opposed stochastic, analytical, and dimension reduction) to deal
to the presence of quantifiable risk. In terms of computa- with aleatory uncertainty and non-probabilistic techniques
tional modeling and simulation perspective, Oberkampf (e.g., interval analysis, convexity approaches) to deal with
et al. (2002) used the term total uncertainty to describe epistemic uncertainty are covered in this section. Output
potential deficiency in any phase or activity of the mod- uncertainty information can be used in reliability-based
eling process. The US National Research Council’s com- design optimization (RBDO) of engineering systems.
mittee on improving risk analysis approaches defined The current manuscript predominantly reviews time- and
uncertainty as a general concept that reflects our lack of space-independent approaches, though we discuss a few
sureness about something or someone, ranging from just time-dependent literature in Sect. 4.1.5. Section 5 provides
short of complete sureness to an almost complete lack of a review of RBDO according to the purpose of research,
conviction about an outcome (Council et al. 2009). It is reliability estimation, and type of uncertainty. Analytical
seen that the content and intent affect the definition of approaches using most probable point (MPP) and sam-
uncertainty in a particular field. pling approaches for RBDO under aleatory uncertainty as
In this paper, we use the definition provided by well as various approaches to deal with RBDO under both
Nikolaidis et al. (2004b), where a useful functional defi- aleatory and epistemic uncertainty induced by insufficient
nition of uncertainty is given as “information/knowledge data are covered in this section. Section 6 reviews bench-
gap between what is known and what needs to be known mark examples and application problems in the literature.
for optimal decisions, with minimal risk.” Based on this Finally, Sect. 7 provides some concluding remarks on the
definition, the uncertainty is classified into two catego- current status of uncertainty treatment practices and pos-
ries: aleatory uncertainty and epistemic uncertainty. Alea- sible future research directions.
tory uncertainty refers to the inherent uncertainty due to
probabilistic variability, and is also known as statistical 2 Fundamentals
uncertainty. This aleatory uncertainty is irreducible and
usually characterized by a probability distribution. On the 2.1 Uncertainty modeling
other hand, epistemic uncertainty stems from the lack of
knowledge such as inadequate understanding of the under- In this section, we aim to provide some basic formulations
lying processes, incomplete knowledge of the phenom- concerning uncertainty modeling, analysis, and optimiza-
ena, or imprecise evaluation of the related characteristics tion. An experienced researcher in uncertainty treatment
(Alleman 2014). Epistemic uncertainty is also known as might choose to skip this section. Commonly used uncer-
systematic uncertainty, and is reducible if we have more tainty handling theories are probability theory (the most
information on the system. Epistemic uncertainty can be used), evidence theory, fuzzy set theory, possibility theory,
characterized by methods such as probability bounds anal- interval analysis, info-gap decision theory, and hybrid
ysis, fuzzy logic, or Dempster–Shafer theory. This paper approaches.
will review uncertainty modeling, uncertainty analysis, In probability theory, first a sample space Ω is defined
and design under uncertainty according to the definition that relates to the set of all possible outcomes. For each
of uncertainty and its categorization explained above. element x ∈ Ω , the probability function f(x) is associated,
which satisfies
∑
1.3 Organization of the paper f (x) ∈ [0, 1] for all x ∈ Ω; and f (x) = 1 (1)
x∈Ω
The remainder of the paper is organized as follows. Sec-
Each subset of Ω is called an event E, and the probability of
tion 2 provides a fundamental background of uncertainty
an event is defined as
treatment—uncertainty modeling, uncertainty analysis,
and design under uncertainty—in engineering applica- ∑
Pr(E) = f (x) (2)
tions. Section 3 presents a review of uncertainty modeling x∈E
practices. Uncertainty categorization (or classification),
uncertainty handling (or management), and uncertainty In evidence theory, first a space of mass is defined as
m ∶ 2X → [0, 1] , where X is the universal set including all

13
Modeling, analysis, and optimization under uncertainties: a review

possible states, and 2X is the set of all the subsets of X. The opportuneness. The robustness strategy satisfices the out-
mass function is also called basic belief assignment. For a come and maximizes the immunity to error, and this strat-
subset S ∈ 2X , m(S) is derived from the evidence that sup- egy is different from outcome optimization. The opportune-
ports S: ness strategy, on the other hand, seeks windfalls at minimal
∑ uncertainty.
m(S) = 1 (3)
S∈2X
2.2 Uncertainty analysis
Evidence from different sources are combined to arrive at a
degree of belief. Belief is the summation of all the evidence In the context of structural design, one usually solves differ-
that fully supports S, and plausibility is the summation of all ent variants of the following equation:
the evidence that partly or fully supports S. That is,
∑ ∑ y = f (𝐱) (7)
belief(S) = m(T), plausibility(S) = m(T) (4)
T⊆S T∩S≠𝜙
where 𝐱 is the vector of input variables which are mostly
independent variables. However, if input variables are
The probability of a set S ∈ 2X falls into the range of dependent, they can be converted to independent vari-
[ belief(S), plausibility(S) ]. In fuzzy set theory, the notion ables through Rosenblatt transformation. y is the output or
of a regular crisp set is extended by introducing a member- response and f is the function that relates 𝐱 and y. In real
ship function. In this theory, there is a gradual rather than life, most of these inputs are uncertain and it is imperative to
sharp transition between non-membership and full mem- design against these uncertainties. As a first step, the uncer-
bership. For each element x ∈ Ω , the membership function tainties in 𝐱 are handled based on the class of uncertainty as
𝜇A ∶ 2X → [0, 1] depicts the degree of membership. The discussed earlier. The characterized uncertainties are then
membership function 𝜇A and the set(A) constitutes a fuzzy propagated through f. f oftentimes is not available explicitly
set. and researchers typically use approximations such as surro-
In possibility theory, two measures are attached to one gates or metamodels to obtain an emulator for f. Establishing
event, namely a ‘necessity measure’ and a ‘possibility meas- f requires an 𝐱 and the corresponding y and typically follows
ure.’ Both are membership functions that can take values the form:
between 0 and 1. If an event A is completely necessary,
its necessity measure is one (N(A) = 1) , and the possibil- y = 𝐰T 𝐱 + v (8)
ity measure of its complement event is zero (Π(A) = 0) .
The weights 𝐰 are essentially the coefficients and one finds
Similarly, if an event is completely possible, its possibil-
them using an optimization formulation on least square
ity measure is one (Π(A) = 1) , and necessity measure of its
or maximum likelihood approaches, while v is a constant.
complement event is zero (N(A) = 0) . To characterize the
Though the regression itself is linear, complex functions can
uncertainty of event A, both of these measures are needed.
be approximated by using different basis functions such as
The necessity degree describes the indications supporting
radial basis and Gaussian process. Obtaining y is usually an
the event, and 1 minus the possibility degree describes the
expensive process and requires a physical experimental set
indications weighing against it.
up or a computational analysis model. Once the surrogate
The interval analysis aims at placing upper and lower
is built and uncertainty is propagated, the resultant y is also
bounds for the range of a function defined in terms of uncer-
random which calls for uncertainty handling approaches
tain variables. Real intervals are typically used as given
such as reliability or robustness-based design to account for
below, where a = −∞ , and b = +∞ are allowed.
the uncertainty in y while making decisions based on the
[a, b] = {x ∈ ℝ|a ≤ x ≤ b} (5) response.
Depending on the amount and type of data available, the
In info-gap decision theory, uncertainty level 𝛼 of a param- approaches used to address the uncertainties are either prob-
eter x is modeled by using the envelope model as abilistic or non-probabilistic in nature. The amount of data
| x − x̃ | that one can seek and the type of data influence the choice of
U(𝛼, x̃ ) = || |≤𝛼
| (6) design of experiment which is the preceding step to model
| x̃ |
building and analysis. Majority of the approaches that pos-
where x̃ is the point estimate of x, and U(𝛼, x̃ ) is the set of all sess an underlying probabilistic essence fall under one or a
values of x whose deviation from x̃ will never be more than combination of the following: sampling, stochastic, analyti-
𝛼̃x . The decision maker does not know the values of x and cal, and dimension reduction techniques. Sampling-based
𝛼 . Two decision concepts are used, namely robustness and approaches are usually variants of the Monte Carlo simula-
tion (MCS) where each realization is propagated through the

13
 E. Acar et al.

model. The model itself is usually a surrogate that can be The primary idea in active subspace is to rotate the coor-
built on one shot design of experiment, adaptively refined, dinates such that the directions of the strongest variation
ensembled, or be multi-fidelity. Once the model is available, are aligned with the rotated coordinates and the model built
crude MCS can be used to estimate probability of an event and analysis performed along the most important rotated
(E) as follows: coordinates. P-box based approaches are used to obtain con-
N
servative estimates when scarce samples are available and
1∑ probabilistic approaches are required.
Pr(E) ≃ I[G(̂xj ≤ 0)] (9)
N j=1 When the data available are scarce and usually in terms
of bounds or intervals, non-probabilistic approaches are
where N is the number of samples, I is an indicator function widely used. When the available information is in the
which takes a value of 1 if [.] is true and 0 if [.] is false. G is form of interval variables, interval approaches are used to
the limit-state function and xj is the jth random realization model the uncertainties and perform optimization. Interval
from PDF fx () . Since evaluating I is usually expensive, one approaches are usually conservative but better than deter-
would like to keep the evaluation count as low as possible. ministic designs. Convex models combine concepts of inter-
Several surrogate models such as noise-free Gaussian pro- val and convexity concepts to develop uncertainty repre-
cess (GP), which is also called Kriging, radial basis function sentation. The idea of convexity approaches is to bound the
(RBF), etc. are widely used. Despite the use of the surrogate uncertain domain using different geometric shapes and use
model, more efficient samplings than the crude MCS are their properties for quantifying uncertainties.
necessary such as importance sampling (IS), line sampling,
and subset simulation. In addition, various sequential sam- 2.3 Design optimization under uncertainties
pling methods such as efficient global reliability analysis
and local adaptive sampling have been proposed to further RBDO is to find a reliable optimum while satisfying proba-
improve the efficiency of the surrogate modeling. bilistic constraints. It has gained wide popularity in engi-
Since estimating small failure probabilities requires large neering applications through accurate reliability analysis
samples, techniques such as IS, subset simulation, and adap- under various uncertainties. Depending on the uncertainty
tive sampling techniques that progressively place samples in it deals with, design optimization under uncertainty is classi-
the regions of interest are developed and used. Polynomial fied into two categories: (1) design optimization under alea-
chaos expansion presented in (10) is one of the widely used tory uncertainty also called RBDO and (2) design optimiza-
stochastic approaches to propagate the uncertainties. tion under both aleatory and epistemic uncertainties. RBDO
p
∑ p
∑ under aleatory uncertainty basically assumes that input sta-
y≈ Ci 𝜑i (x) = Ci Πdj=1 Pj (xj ) (10) tistical models and limit-state functions are fully known and
i=1 i=1 perfectly accurate. On the other hand, when epistemic uncer-
tainty caused by lack of knowledge and insufficient data is
where p is the number of coefficients, Ci is the expan-
involved in the design optimization, it is extremely difficult
sion coefficients, 𝜑 is the multivariate polynomial that is
to know exact input statistical models and output simulation
obtained as a product of d-univariate orthogonal polynomi-
models in the real world.
als Pj which allows use of powerful statistical properties.
In general, RBDO is formulated as
There are also developments in stochastic collocation type
of approaches. Popular analytical approaches include first- min cost(𝐝, 𝜇𝐗 , 𝜇𝐏 )
order reliability method (FORM), second-order reliability
𝐝,𝜇𝐗

method (SORM), and dimension reduction method (DRM) subject to Pr(G(𝐝, 𝐗, 𝐏) ≤ 0) ≥ Retarget
according to how a limit-state function is approximated.
where Pr(G(𝐝, 𝐗, 𝐏) ≤ 0) ≡ f𝐱,𝐩 (𝐗, 𝐏)d𝐱d𝐩
Major inroads were made in contributing to the reliability �G(𝐝,𝐗,𝐏)≤0
estimates such as probabilistic performance measure, prob- (11)
abilistic sufficiency factor, and percentile measure. These
where G is the limit-state function, 𝐝 is deterministic design
inverse measures operate in performance space and aid in
variable, 𝐗 is the random variable, 𝐏 is the random parame-
stable convergence. When the dimensions are large, even
ter, and Retarget is the target reliability. The probability meas-
surrogates suffer from the curse of dimensionality. Hence,
ure Pr in (11) is equivalent to a multidimensional integration
dimension reduction approaches are preferred. The underly-
where f𝐱,𝐩 (𝐗, 𝐏) is a joint probability density function (PDF)
ing idea is to combine the dimensions in a linear sense along
of random variables and parameters
a dimension of larger variation. While approaches such as
principal component analysis were used in the past, recent
techniques such as active subspace are being widely adopted.

13
Modeling, analysis, and optimization under uncertainties: a review

2.3.1 Design optimization under aleatory uncertainty with experimental results. It is necessary to calibrate the
simulation model to estimate unknown model parameters
The multidimensional integration in (11) is difficult or even and discrepancy and validate it. Thus, model calibration and
impossible to accurately compute. Hence, RBDO under validation methods such as Bayesian model calibration and
aleatory uncertainty approximates either the limit-state optimization-based model calibration have been studied.
function (analytical approach) or the numerical integration The seminal work, called KOH framework (Kennedy and
(sampling approach). To alleviate the computational cost O’Hagan 2001; Roy and Oberkampf 2011), described vari-
for the double-loop RBDO in (11), decoupled loop such as ous uncertainties in simulation model and gives a Bayes-
sequential optimization and reliability assessment (SORA) ian approach to deal with the uncertainties, especially for
and single loop such as single-loop approach and single- model calibration, and thus it may be helpful to understand
loop single vector methods have been proposed where the the model uncertainties. Even though the simulation model
probabilistic constraints can be transformed to deterministic is calibrated and validated, simulation model uncertainty
constraints. To alleviate the computational cost for the sam- cannot be perfectly eliminated since experimental data may
pling approach, efficient and accurate surrogate modeling be limited in the real world. Thus, the simulation model
methods and various sequential sampling methods have been uncertainty has to be quantified and taken into account in
proposed. design optimization.

2.3.2 Design optimization under both aleatory

and epistemic uncertainties 3 Uncertainty modeling

In this study, epistemic uncertainty in design optimization is Uncertainty modeling constitutes uncertainty categorization
categorized into input and output model uncertainties. The (or classification), uncertainty handling (or management),
purpose of design optimization under epistemic uncertainty and uncertainty characterization (Fig. 1). Uncertainty cat-
is not to reduce the epistemic uncertainty but quantify it for egorization refers to the classification of uncertainty into
reliability estimation. There are two ways of dealing with the different categories. This classification is an important step
epistemic uncertainty in the input statistical model: (1) non- before uncertainty handling and quantification because spe-
probabilistic approach and (2) probabilistic approach. In the cific methods on uncertainty handling and quantification are
non-probabilistic approach, the typical RBDO formulation in suitable for specific class of uncertainty. Uncertainty han-
(11) may not be applicable since the input statistical model dling refers to the management of uncertainties by various
is not described as PDF. Instead, interval analysis, evidence theories such as probability theory, evidence theory, fuzzy
theory, possibility theory, and fuzzy set theory described set theory, possibility theory, interval analysis, and info-gap
using membership functions can be used to describe the decision theory. Uncertainty characterization refers to statis-
epistemic uncertainty. Using these, mixed-variable design tical description of input uncertainties. It includes distribu-
optimization where interval and random variables coexist, tion fitting (e.g., PDF, membership function), parameter esti-
evidence-based design optimization, and non-probabilistic mation (e.g., mean value, standard deviation, quantile), and
RBDO have been proposed. On the contrary, the probabilis- correlation modeling of the input uncertainties. Note that
tic approach attempts to describe the incomplete input statis- uncertainty characterization is the first stage of uncertainty
tical model using non-parametric kernel density estimation quantification (UQ) that also involves uncertainty analysis
(KDE), the Bayesian approach combined with the bootstrap- and propagation, which will be covered in Sect. 4.
ping method, and parametric distributions with conservative
parameter estimation. In this case, confidence-based design 3.1 Uncertainty categorization (or classification)
optimization utilizes the confidence of reliability to replace
the probabilistic constraints in (11). Uncertainty is usually classified into aleatory uncertainty
Output models, such as surrogate models or simulation and epistemic uncertainty (Hoffman and Hammonds 1994;
models, may not perfectly emulate simulation models or Rowe 1994; Hora 1996; Ferson and Ginzburg 1996; Paté-
experiments, respectively, and it can lead to inaccurate reli- Cornell 1996; Ferson et al. 2004; Acar et al. 2006; Sanka-
ability analysis. Thus, design optimization under surrogate raraman and Mahadevan 2011; Li et  al. 2012). In this
model uncertainty has been proposed, which utilizes the pre- distinction, epistemic uncertainty includes both the non-
diction variance calculated from a Kriging model to obtain deterministic behavior due to the lack of knowledge (e.g.,
the distribution of reliability. In other words, the uncer- mathematical modeling approximations), and also the recog-
tainty induced by an inaccurate surrogate model is taken nizable deficiency that is not due to lack of knowledge (e.g.,
into account to prevent the overestimation of reliability. On computer programming errors).
the other hand, the simulation model always has discrepancy

13
 E. Acar et al.

Uncertainty modeling

Uncertainty categorization (or Uncertainty handling (or Uncertainty

classification) management) characterization

Aleatory Probability theory Parametric approach in

probabilistic handling

Epistemic Evidence theory

Non-parametric approach in
Fuzzy set theory probabilistic handling

Possibility theory Characterization in non-

probabilistic handling
Interval analysis

Info-gap theory

Hybrid approaches

Fig. 1  Uncertainty modeling tree structure

Aleatory uncertainty also referred to in the literature 3.2 Uncertainty handling (or management)
as variability, stochastic uncertainty, inherent uncertainty,
and irreducible uncertainty is recognized as the inherent The main theories used for uncertainty handling (or manage-
randomness originating from the natural variability of the ment) can be considered as the following: (1) probability
physical system. Aleatory uncertainty cannot be eliminated theory, (2) evidence theory, (3) fuzzy set theory, (4) possibil-
or reduced by collecting more information or gathering more ity theory, (5) interval analysis, (6) info-gap decision theory,
knowledge. Epistemic uncertainty also referred to in the lit- and (7) hybrid approaches. The main difference among these
erature as subjective uncertainty, informative uncertainty, approaches relate to the techniques used for describing the
and reducible uncertainty is recognized as non-deterministic uncertainty in input parameters. The origins of different
behavior due to lack of knowledge of the physical system approaches are presented in this section, and advances on
along with the ability of modeling and measuring the physi- these approaches are discussed in the next section.
cal system. Unlike aleatoric uncertainty, epistemic uncer-
tainty can be reduced through quality control (Acar et al. 3.2.1 Probability theory
2007), structural testing (Acar et al. 2010), non-destructive
inspection (Kale and Haftka 2008), and sometimes can even Probability theory is the oldest and the most widely used
be eliminated. uncertainty handling theory. Parsons and Hunter (1998) note
There also exist studies that use more than two classes for that this theory dates back to several hundred years and it is
uncertainty categorization. For instance, Oberkampf et al. difficult to state where the definitive account may be found.
(2002) classified uncertainty as variability, uncertainty, and The earliest known forms of probability and statistics were
error. In that classification, variability describes the inherent developed by Arab mathematicians studying cryptography
variation associated with the physical system under consid- in the eight century according to Broemeling (2011). The
eration. Uncertainty is defined as a potential deficiency in classical interpretation is known to be completed by Laplace
any phase or activity of the modeling process that is due to (1812) according to Hájek (2019), in probability theory, it is
lack of knowledge. Error is defined as a recognizable defi- assumed that the input parameters are random variables with
ciency in any phase or activity of modeling and simulation a known PDF or cumulative distribution function.
that is not due to lack of knowledge.

13
Modeling, analysis, and optimization under uncertainties: a review

In structural mechanics, geometric parameters, loading, 3.2.2 Evidence theory

and material properties are random and are usually repre-
sented by PDF or cumulative distribution function. Most The evidence theory was first developed by Dempster (1967)
manufacturing operations produce normally distributed and extended by Shafer (1976). Therefore, this theory is also
dimensions if they are performed in a controlled manner called the Dempster–Shafer theory. This theory combines
(Rao 1992). The dead loads (e.g., gravity load) are often evidence from different sources and arrives at a degree of
represented with normal distribution (with 10 % coefficient belief by taking into account all the available evidence. In
of variation), and the live loads are usually modeled with this theory, belief and plausibility are defined as the lower
extreme type I distribution (Ellingwood 1980). It is advised and upper boundary, and a set of belief and plausibility dis-
that the type-II extreme value (or Frechet) distribution could tribution functions is used to describe the input uncertainty.
be used to describe the yearly maximum wind velocity at any While handling subjective uncertainty arising from
location (Thom 1960). experts, Bayes theorem requires prior and error assumptions,
Hess et al. (2002) analyzed the data bank generated by and the obtained results are sensitive to these assumptions.
Kaufman and Prager (1990) by using the computer program Evidence theory, on the other hand, does not require these
BestFitⓇ (1995) to explore which PDF were most represent- assumptions. Soundappan et al. (2004) provides a compari-
ative of the sample data for the material properties of steel son of Bayesian approach and evidence theory in handling
used in marine applications. They found that lognormal, epistemic uncertainty.
Weibull, and extreme value distributions were good choices
for describing the yield strength of steel, whereas Weibull 3.2.3 Fuzzy set theory
and normal distributions well represented the ultimate
strength as well as the elastic modulus of steels. There also Fuzzy set theory is developed by Zadeh (1965) and it models
exists studies on probabilistic modeling of fatigue endurance epistemic uncertainty through fuzzy sets with membership
of steel. Mischke (1987) found that the fatigue life data of functions. Fuzzy set theory extends the notion of a regular
steel alloys can be well represented with Weibull distribution crisp set and expresses classes with vague boundaries such
and lognormal distribution. as tall, good, and important. It provides a natural way of
Bayes theorem is based on probability theory (Bayes dealing with problems in which the source of uncertainty is
1991). It relates two or more events through conditional the absence of sharply defined criteria of class membership
probabilities to make inferences. It is often used when there rather than the presence of random variables (Zadeh 1973).
is a lack of direct information about an event. It describes In this theory, the transition between non-membership and
the probability of an event, based on prior knowledge and full membership is gradual rather than sharp, and a fuzzy set
provides a way to update the probability of that event (poste- is represented by stating its membership function, where a
rior probability) given new or additional evidence. Jiao and degree of membership in the interval [0,1] is given to every
Moan (1990) used Bayesian theory to investigate the effect element in the set (Zimmermann 2001).
of proof tests on structural safety. Beck and Katafygiotis
(1998) addressed the problem of updating a probabilistic 3.2.4 Possibility theory
structural model using dynamic test data from structure by
using Bayes theorem. Acar et al. (2010) used Bayes theo- Possibility theory is based on the notion of fuzzy sets, and it
rem to update the failure stress distribution of an aircraft was first presented by Zadeh (1978). It utilizes two measures
structural element based on results of the element tests. It is attached to one event, namely a “necessity measure” and a
also worthy to note that a probabilistic graphical model enti- “possibility measure.” Both are membership functions that
tled Bayesian network (Mahadevan et al. 2001; Pearl 2014) can take values between 0 and 1 (Dubois and Prade 1988).
is established to symbolize the random variables and their Possibility theory is analogous to probability theory as
conditional dependencies by using a probabilistic directed they both use the [0,1] interval for their measures as the
acyclic graphical model. range of their respective functions (Zimmermann 2001). The
In a probabilistic approach, probability distribution is fundamental difference between probability and possibility
used to characterize uncertainty. In some cases (e.g., scarce is that probability is a measure of the frequency of occur-
data), it may not be possible to specify the precise values of rence of an event, while possibility is used to quantify the
the input distribution parameters, precise probability dis- meaning of an event (Agarwal and Nayal 2015). Probability
tributions, and dependencies between input parameters. In distribution functions are required to add to 1, while for pos-
these cases, probabilistic approach is not effective, and other sibility distributions the largest values are required to be 1.
uncertainty handling theories are developed to address some Therefore, possibility can be seen as an upper probability.
of these limitations.

13
 E. Acar et al.

3.2.5 Interval analysis et al. (2001) combined probabilistic, fuzzy, and anti-optimi-

zation approaches; Guyonnet et al. (2003) combined prob-
Roots of interval analysis can be traced back to Moore abilistic and possibilistic approaches; Jiang et al. (2012a)
(1966). It is a branch of the numerical analysis that allows us combined probabilistic and interval analysis approaches;
to compute closed intervals for the exact values of integrals. Chutia (2017) combined probabilistic and fuzzy approaches.
In interval analysis, the uncertain parameter is denoted by a Jiang et al. (2017) present a literature review on probability-
simple range, and if a preference function is to describe the interval hybrid uncertainty analysis for structures with both
desirability of using different values in this range, then fuzzy aleatory and epistemic uncertainties.
theory can be used (Rao and Berke 1997).
When the uncertainties in structural system are modeled
3.3 Uncertainty characterization
with random variables, the failure probability and reliability
index are deterministic values. However, the failure proba-
Uncertainty is characterized by using a probability distribu-
bility (or reliability index) becomes an interval number with
tion function (e.g., PDF) when it is handled by probabil-
the lower and upper bounds if interval variables are included
ity theory. While using probability theory, a parametric or
in the structural system (Gao et al. 2011).
a non-parametric approach can be used (McFarland and
Convex method (Elishakoff et al. 1994a), anti-optimiza-
Mahadevan 2008; Kim et al. 2019). Brief details of these
tion method (Elishakoff et al. 1994b), perturbation method
approaches are given below, followed by the studies on cor-
(Chen and Yang 2000), interval finite element method
relation modeling. Finally, uncertainty characterization in
(Muhanna et al. 2005), MCS method (Sim et al. 2007), and
non-probabilistic uncertainty handling approaches is pre-
affine arithmetic (Degrauwe et al. 2010) have been combined
sented in this sub-section.
with interval operations to analyze structures with interval
parameters.
3.3.1 Parametric approach in probabilistic handling
3.2.6 Info‑gap decision theory
Parametric approach consists of two steps. In the first step,
Info-gap decision theory is developed by Ben-Haim (2001). a proper distribution function is determined (Kang et al.
It is a non-probabilistic decision theory for prioritizing 2019b), which could be performed based on expert knowl-
alternatives and making choices and decisions under deep edge (Soundappan et al. 2004) or by using a model selec-
uncertainty (Ben-Haim 2006). An “info-gap” is the dispar- tion method such as Akaike information criterion (Akaike
ity between what is known and what needs to be known 1974), Bayesian information criterion (Burnham and Ander-
for a responsible decision. Info-gap models of uncertainty son 2004), and Bayesian method (Schwarz 1978; Noh et al.
represent uncertainty in parameters and in the shapes of 2010). In the second step, the parameters of the distribution
functional relationships. function are estimated by using a goodness of fit test such as
Info-gap decision theory offers two decision concepts: Anderson–Darling test (Anderson and Darling 1952), Kol-
robustness and opportuneness. The robustness strategy sat- mogorov–Smirnov test (Kolmogorov 1933; Smirnoff 1939;
isfices the outcome and maximizes the immunity to error, Kolmogoroff 1941), or chi-squared test (Ayyub and McCuen
and this strategy is different from outcome optimization. The 2016). The main downside of the parametric approach is
opportuneness strategy, on the other hand, seeks windfalls at that the use of incorrect distribution function may lead to
minimal uncertainty. Info-gap decision theory has been used erroneous results.
in truss optimization, (Kanno and Takewaki 2006), pipeline In many structural problems, the input random variables
reliability improvement (Cicala and Irias 2014), energy hub are correlated (Annis 2004; Nikolaidis et al. 2004a). How-
management of electric vehicles (Soroudi and Keane 2015), ever, these variables have often been assumed to be inde-
wind power uncertainty analysis (Soroudi et al. 2017), air- pendent because of the difficulty in constructing the joint
plane landing gear design (Platz and Götz 2017), freshwa- distribution of correlated input variables (Noh et al. 2010).
ter management in coastal aquifers (Ranjbar and Mahjouri Even when the correlation is taken into account, usually the
2019), etc. joint Gaussian distribution has been used while the correct
joint distribution could be non-Gaussian (Nataf 1962; Noh
3.2.7 Hybrid approaches et al. 2009). In that case, copulas (functions that couple
multivariate distribution functions to their one-dimensional
Different theories used for uncertainty handling can be com- marginal distribution functions) can be used (Noh et al.
bined to take advantage of the ability of each theory. Ferson 2010). Similarly, intrusive (Paulson et al. 2017) or non-
and Ginzburg (1996) combined probability theory and inter- intrusive (Lin et al. 2020) polynomial chaos method can be
val analysis to produce probability boxes (p-boxes); Tonon used to deal with correlated random variables, where the

13
Modeling, analysis, and optimization under uncertainties: a review

multivariate orthogonal polynomial basis corresponding to on the fact that a larger entropy of an information system
the correlated input random variables is constructed by solv- indicates more information contained in the system (Martin
ing the moment-matching equations based on the correlation and England 1981). Civanlar and Trussell (1986) presented
statistical moments. In some problems, point and interval a guideline to construct the membership functions for fuzzy
samples might be available for the estimation of distribu- sets whose elements have a defining feature with a known
tion parameters. For those problems, the correlations among PDF, and showed that their method is capable of generat-
interval distribution parameters can be modeled using ellipse ing membership functions in accordance with the possibil-
models (Xiao et al. 2020). ity–probability consistency principle. Jang (1993) presented
an adaptive-network-based fuzzy inference system, a fuzzy
3.3.2 Non‑parametric approach in probabilistic handling inference system implemented in the framework of adaptive
networks. By using a hybrid learning procedure, adaptive-
In non-parametric approach, on the other hand, probabil- network-based fuzzy inference system can construct an
ity distribution is determined directly from the data. Histo- input–output mapping based on both human knowledge (in
grams or KDE are widely used in non-parametric approach the form of fuzzy if-then rules) and stipulated input–output
(McFarland and Mahadevan 2008; Cho et al. (2016c)). The data pairs. Hong and Lee (1996) proposed a general learning
non-parametric approach is recommended over parametric method as a framework for automatically deriving member-
approach if the random variables follow non-parametric dis- ship functions and fuzzy if-then rules from a set of given
tributions or the number of given data is insufficient, even training examples to rapidly build a prototype fuzzy expert
though the true distribution of the data is a parametric distri- system. Simon (2002) presented a modified form of gradi-
bution (Kang et al. 2017b). However, if the number of data ent descent and Kalman filter methods for optimization of
is very small (e.g., less than 10 samples), the non-parametric asymmetric triangular membership functions. Determination
approach is very sensitive to the quality of the given data of fuzzy membership functions through genetic algorithm
and it may lead to erroneous results (Kang et al. 2019a). (Arslan and Kaya 2001), ant colony optimization (Jiang et al.
Non-parametric approach can be combined with an interval 2008), and particle swarm optimization (Omizegba and Ade-
analysis to overcome the limitations of the non-parametric bayo 2009) was also presented. Hasuike and Katagiri (2016)
approach (Kang et al. 2018). constructed the appropriate membership function based on
For the correlation modeling for non-parametric size of fuzzy set and mathematical programming. Jalota
approaches, multivariate non-parametric KDE is often used et al. (2017) constructed membership function for uncer-
(Wang and Wang 2015a). The KDE is similar to the use tain portfolio parameters by using a credibilistic framework.
of empirical probability mass function, but each point of
the mass function is replaced with a continuous, symmet-
ric distribution centered at that point. The scale parameter 4 Uncertainty analysis
of the symmetric distribution (i.e., the bandwidth) has a
substantial effect on the performance of the KDE. Ahmad In this section, we discuss the uncertainty analysis mod-
(1982), Wand and Jones (1994), and Duong and Hazelton els, methods, and tools. The quantified uncertainties using
(2003) used fixed bandwidth, whereas Zhang (2011) and uncertainty characterization or modeling techniques that
Zougab et al. (2014) used adaptive (or variable) bandwidth were discussed in the previous section need to be propagated
in KDE. In univariate case, the bandwidth is a scalar, where through the analysis models to obtain random responses or
it turns into a matrix (e.g., the covariance matrix). The outputs. The resulting uncertain output needs to be quanti-
selection of the bandwidth or the covariance matrix can fied and is commonly referred to as model uncertainty analy-
be done through cross-validation (Bowman 1984; Duong sis (Ghanem et al. 2017). Based on system complexities,
and Hazelton (2005)), maximum likelihood (Wang 2007; the analysis models could be simple closed-form analytical
Konečná and Horová 2019), Bayesian approach (Zhang et al. functions or computer models such as finite element mod-
2006; Zougab et al. 2014), or method of penalizing functions els, computational fluid dynamic models, or physical experi-
(Bashtannyk and Hyndman 2001). ments. Uncertainty analysis involves accessing these analy-
sis models repeatedly to obtain a random characterization
3.3.3 Uncertainty characterization in non‑probabilistic of the output. That is, in a sampling perspective, for each
handling realization of the random input, the analysis model needs to
be accessed to obtain the corresponding response. The inter-
Uncertainty is characterized by using a membership func- est in the output includes finding the moments, estimating
tion when it is handled by fuzzy set theory. Cheng and Chen tail probabilities, and computing the PDFs towards design
(1997) determined the membership function such that the under uncertainty such as robust or reliability-based design.
corresponding fuzzy event has maximum entropy, based

13
 E. Acar et al.

Uncertainty analysis

Probabilistic Non-probabilistic
Monte Carlo

Sampling Importance sampling Interval analysis

Adaptive / Sequential
Convexity approaches
Taylor series

Stochastic or functional Perturbation methods

and local expansion Fuzzy, Possibility theory

Polynomial chaos expansion

Analytical FORM / SORM

Eigenvector dimension
reduction
Dimension reduction

Active subspace

Fig. 2  Uncertainty analysis tree structure

The uncertainty analysis tree structure is presented in 4.1.1 Sampling‑based approaches

Fig. 2. A wide variety of uncertainty analysis approaches
can be broadly classified into (1) probabilistic and (2) non- Sampling-based approaches predominantly include MCS
probabilistic techniques. Depending on the availability of techniques (Madsen et al. 2006) or its variants such as sepa-
probabilistic information, output requirement, and complex- rable Monte Carlo (Smarslok et al. 2010), IS (Cadini et al.
ity of the analysis model, either of the above techniques can 2015), multilevel Monte Carlo approaches (Giles 2008), and
be used. Any of the techniques discussed in the subsections adaptive sampling techniques (Liu et al. 2018b). In the event
below can be used on the models directly or on metamodels of knowing the probability information of the inputs, MCS
constructed based on design of experiment points. is directly used in the crude form if the analysis model is
computationally cheap to evaluate. Oftentimes, the analysis
4.1 Probabilistic techniques model is replaced by a metamodel (Li et al. 2012). Popu-
lar metamodels include polynomials (Hosder et al. 2001),
There are different approaches that fall under the probabil- Kriging (Booker et  al. 1999), support vector regression
istic techniques umbrella (Lee and Chen 2009). Required (SVR) (Basudhar and Missoum 2008), or their ensembles
probabilistic information can be modeled using differ- (Goel et al. 2007; Viana et al. 2009). Recent developments
ent approaches as discussed in the previous section. The in metamodels for uncertainty treatment include approaches
approaches in probabilistic techniques for uncertainty anal- such as deep GP (Radaideh and Kozlowski 2020) and game
ysis can be broadly classified into 4 categories as shown theory-driven approaches (Chen et al. 2020). Basudhar and
in Fig.  2. All categories are discussed in the following Missoum (2008) and Basudhar et al. (2008) applied sup-
subsections. port vector machine (SVM) for solving limit-states with
discontinuous limit-states as well. In large variable prob-
lems, beyond the usual screening process to identify influ-
ential inputs, Iooss and Le Gratiet (2019) have suggested to
model the non-influential inputs as another GP. Using the

13
Modeling, analysis, and optimization under uncertainties: a review

joint metamodel, the uncertainties are then propagated. Long failures as less as 10−6 while achieving an order of magni-
run times in end-to-end complex designs with uncertainties tude reduction in the number of required runtime. A similar
are avoided by using a combination of grouped sensitivity approach is proposed by Echard et al. (2013) while using an
analysis, expert opinions to certify surrogate models, and active Kriging metamodel and by Cadini et al. (2014) for
verification and validation techniques (Allaire et al. 2014). estimating multiple low probability failure regions. Yang
In the context of multidisciplinary problems, Baudoui et al. et al. (2018) combined active Kriging with IS for low failure
(2012) suggest processing the uncertainty locally in each probability estimates. Zhang et al. (2020) proposed adaptive
discipline with minor changes to the initial multidisciplinary Kriging and IS for system reliability analysis.
system. Hoseyni et al. (2014) use a hybrid qualitative and In addition to IS, other adaptive sampling techniques
quantitative approach for uncertainty assessment. It is to be which usually focus on building accurate metamodels locally
noted that there are other reduced order modeling techniques are also used for uncertainty analysis. Volpi et al. (2015)
such as projection-based approaches (Benner et al. 2015), develop a dynamic RBF approach where the stochastic ker-
Krylov spaces (Freund 2003), and Centroidal voronoi tes- nel is defined by an uncertain tuning parameter. The effect
sellations which are typically used in applications such as of tuning parameters on prediction are determined by UQ
computational fluid dynamics, control theory, and dynamics. methods. Prediction uncertainty and parallel infill are used
While literature discussed above employs models of par- for adaptive sampling and effectively reduce the number of
ticular fidelity throughout the analysis, there are also tech- high-fidelity simulations. Weinmeister et al. (2018) com-
niques that use multi-fidelity approaches. Ng and Willcox bine polynomial chaos expansion (PCE), Kriging and adap-
(2014) discuss strategies for uncertainty propagation with tive sampling for UQ. While most of the adaptive sampling
variable fidelity models. Peherstorfer et al. (2018) discuss schemes operate on the idea of prediction variance from
several variable fidelity approaches for uncertainty analysis the constructed metamodel, Romero et al. (2004) proposed
and optimization. to use progressive lattice sampling to incrementally add
Since failure probabilities are typically small, MCS samples for accurate metamodels. A similar approach was
requires many model evaluations until samples in the tails discussed by Helton and Davis (2003). Metamodel-based
of the distributions are sampled for an accurate estimate. IS adaptive sampling was proposed by Bichon et al. (2008)
is a variance reduction technique that permits reduction of using efficient global optimization procedure which ensures
number of model evaluations by utilizing a targeted sam- accuracy of the limit-state. Another suite of algorithms focus
pling strategy. Dubourg et al. (2013) use IS along with meta- on estimating the sensitivities of the probability estimates
models to estimate the augmented probability. They then add (Jensen et al. 2009; Valdebenito and Schuëller 2010). Tafla-
a correction term to remove bias in the event of the meta- nidis and Beck (2008a, 2008b) propose the stochastic subset
model not being accurate and demonstrate the approach’s optimization approach where the subset simulation approach
efficiency up to 100 variables. Chaudhuri and Haftka (2013) (Au and Beck 2001) is employed to explore both uncertain
use separable MCS and IS for reliability estimation. Zhang parameter space and design variable space simultaneously.
and Taflanidis (2019) use Kriging to approximate the func- The fundamental idea behind subset simulation is to express
tion in both the design variable and random variable space the failure probability as a product of conditional probabili-
so that UQ and Pareto front can be achieved simultaneously. ties which are usually larger and can be estimated with lesser
They combine the metamodel with IS to sample only in the samples. Subset simulation and its extensions were exten-
regions of interest and selectively propagate the uncertain- sively used to estimate small failure probabilities (Zuev et al.
ties. Peherstorfer et al. (2016) propose a multi-fidelity IS 2012; Meng et al. 2015a; Papaioannou et al. 2015; Šehić and
method where they build a surrogate model for the high- Karamehmedović 2020). Li and Cao (2016) present Matlab
fidelity simulations during the construction of the biasing codes for the algorithm. However, there are limitations as
distribution. Then, a high-fidelity model is evaluated at the well to this approach as pointed by Breitung (2019).
samples drawn from the biasing IS distribution that allows
to estimate small failure probabilities. Papaioannou et al. 4.1.2 Stochastic or functional expansion and local
(2015) propose a sequential importance sampling approach expansion approaches
for estimating reliability. Here, the samples from the ran-
dom variable distributions are translated to samples from an Spectral finite element and approaches such as PCE can be
approximately optimal IS density. Sample transition happens classified under this category. PCE is based on representing
through a sequence of intermediate distributions based on a random variable by a series of polynomial chaos basis.
a resample-move scheme. This is further used (Papaioan- Chaos here refers to the Gaussian random process. The gen-
nou et al. 2018) for computing the sensitivity of reliability eralized PCE generates an orthogonal basis based on dis-
estimates. Cadini et al. (2015) combine metamodels and tribution characteristics of input random variables. Care-
IS approach with an adaptive sampling scheme to estimate fully chosen orthogonal basis yields approximate structural

13
 E. Acar et al.

response after multiple iterations. PCE can also be viewed functions and the distribution of input parameters. These
as a metamodel because the coefficients need to be com- approaches are broadly classified as single-loop, double-
puted. Widely used approaches for the coefficient estima- loop, and decoupled approaches (Valdebenito and Schuëller
tion are least squares and projection approaches. However, 2010; Aoues and Chateauneuf 2010; Bichon et al. 2008).
neither of the approaches escape the curse of dimensional- The double-loop approach has the optimization outer loop
ity. In order to reduce the computational effort when the and reliability estimation inner loop. In the inner loop, reli-
dimensions increase, a series of sparce PCE approaches ability estimates such as Reliability Index (Tu et al. 2001),
are proposed (Blatman 2009; Blatman and Sudret 2010a, probabilistic sufficiency factor (Ramu et al. 2006), proba-
b; Chen et al. 2018; Xu and Wang 2019). The sparse PCE bilistic performance measure (Youn et al. 2003), or percen-
techniques help in reducing the number of expensive com- tile performance measure (Du et al. 2004) are estimated.
putations. Researchers have combined dimension reduction All these approaches focus on estimating the MPP in a for-
technique and sparse PCE to emulate the reduced output ward reliability estimation or inverse reliability approach.
(Nagel et al. 2020) while sensitivity information is obtained Detailed discussions are presented in Valdebenito and
by postprocessing the coefficients. Zhou et al. (2019b) use Schuëller (2010).
sparse PCE for sensitivity analysis. Rajabi (2019) compares There have been numerous studies on efficient strate-
PCE and GP emulation in the context of ground water appli- gies to deal with a non-linear high-dimensional limit-state
cations. They observe that PCE provides better accuracy in function for MPP search in RBDO. In the beginning, there
moment and tail estimates with less variance across different were two different approaches for RBDO: reliability index
runs. Researchers develop a metamodel by combining PCE, approach (Hasofer and Lind 1974) and performance meas-
dimension reduction approach, and information-theoretic ure approach (PMA) (Tu et al. 1999). Several developments
entropy (He et al. 2020). A multi-fidelity PC approach is have been proposed for reliability index approach: Santosh
developed by extending the multilevel co-Kriging multi- et al. (2006) applied the Armijo rule to optimal step length
fidelity modeling framework in the deterministic domain to selection for the Hasofer–Lind and Rackwitz–Fiessler
the stochastic one (Wang et al. 2019a). Zhang et al. (2019) method, and the HLRF–BFGS optimization algorithm was
obtain the multimodal input distributions through a GP and proposed exploiting the BFGS updates to approximate the
propagate them through the response using an integration Hessian matrix since the conventional Hasofer–Lind and
of the sparse grid numerical method and maximum entropy Rackwitz–Fiessler algorithm can be treated as the sequential
method. Wu et al. (2017) use a sparse grid collocation tech- quadratic programming method where the Hessian matrix of
nique to build metamodel and a combination of Bayesian Lagrangian is approximated by an identity matrix (Periçaro
analysis and Markov chain Monte Carlo is used to solve the et al. 2015). In addition, stability transformation method for
inverse UQ problem. Teckentrup et al. (2015) suggest a mul- chaos control (CC) of MPP search using Hasofer–Lind and
tilevel stochastic collocation approach to deal with random Rackwitz–Fiessler is proposed by Yang (2010), In PMA,
inputs. Proper orthogonal decomposition is combined with Youn et al. (2003) proposed the hybrid mean value method
PCE for efficient UQ of complex acoustic wave problems integrating conventional advanced mean value method with
with large number of output physical variables (El Moçayd conjugated mean value method to exploit both methods
et al. 2020). Kumar et al. (2020) use PCE on complex geo- selectively according to whether a limit-state function is
metrically irregular spatial domains and the surrogate mod- convex or concave. In addition, several strategies for effi-
els are constructed using stochastic collocation. The local cient RBDO are developed in enriched performance meas-
expansion-based methods include approaches such as Taylor ure approach (PMA+) (Youn et al. 2005a). A hybrid chaos
series or perturbation method (Der Kiureghian 1996; Gha- control for PMA is proposed by integrating advanced mean
nem and Spanos 1991). Such methods suffer under large value and modified chaos control procedures to find MPP
variability of inputs and non-linearity of performance func- more efficiently and robustly (Meng et al. 2015b). Modified
tions (Lee and Chen 2009). While using the advanced first- chaos control further improves the convergence by extending
order second moment approach to estimate reliability, the the iterative point of CC method to the constraint bound-
suboptimization process to estimate MPP is accelerated by ary (Yang et al. 2020). Keshtegar and Hao (2017) proposed
using the Neumann expansion technique (Lee and Kwak hybrid self-adjusted mean value (SMV) method to enhance
1995) to deal with random state equations. the SMV method. Jung et al. (2020b) proposed an intelligent
initial point for MPP search in RBDO to improve the effi-
4.1.3 Analytical approaches ciency of MPP search utilizing the sensitivity of MPP with
respect to design point.
Analytical approaches such as FORM and SORM approxi- There has been a multiple MPP issue when the limit-state
mate the reliability estimation and allow for analytical function is highly non-linear. To resolve this issue, Der Kiu-
uncertainty analysis. They typically require the limit-state reghian and Dakessian (1998) firstly developed an algorithm

13
Modeling, analysis, and optimization under uncertainties: a review

to find multiple MPPs (i.e., multiple design points), and 4.1.5 Conservative estimates, p‑box models,
Au et al. (1999) tried to resolve the issue using asymptotic and time‑dependent reliability models
approximation and IS. However, research on multiple MPPs
has been still limited so far. When the available sample or probabilistic information
is less, conservative estimates are also used by research-
4.1.4 Dimension reduction approaches ers (Picheny et al. 2008; Cho et al. 2016a; Ito et al. 2018).
Surrogates are also constructed and used in a conservative
When the number of variables is large, DRM is a sought estimate perspective (Viana et al. 2010; Zhao et al. 2013).
after technique. The underlying idea of most DRM tech- Iooss and Le Gratiet (2019) approximate the functional risk
niques is to translate the function from regular variable space curve using a metamodel and provide the confidence due
to reduced dimension variable space. The variables in the to the approximation. They use the perturbed-law based
reduced dimension are usually a combination of the regular sensitivity indices to understand the effect of misjudgment
variables that contribute the most. Mean-based DRM (Rah- on the sensitivity of the functional risk curve to the input
man and Xu 2004; Xu and Rahman 2004; Lee et al. 2008a) parameter’s PDF.
and MPP-based DRM (Rahman and Wei 2006) were pro- When the distribution type or moment information is
posed and widely used. Lee et al. (2008b) use the inverse known, several approaches are available to develop the dis-
MPP-based approach to enable efficient sensitivity estima- tributional p-box model (Zhang et al. 2010a; Oberguggen-
tion. Eigenvector dimension reduction (EDR) method for berger and Fellin 2008; Lee et al. 2016; Wang et al. 2018).
probabilistic analysis without sensitivity information was When the distribution or moment information are not avail-
proposed (Youn and Wang 2008; Youn and Xi 2009). In able, Liu et al. (2019) construct the cumulative distribution
order to overcome the limitations of the univariate approxi- function p-box based on the maximum entropy principle.
mation (Lee et al. 2008b), several advances have been imple- The interval Monte Carlo is developed by combining the
mented (Bae and Alyanak 2016; Kang et al. 2017a; Jung interval sampling and interval finite element method for
et al. 2019a; Park et al. 2020). Researchers combine a DR uncertain analysis with p-boxes (Zhang et  al. 2010a, b,
technique and maximum entropy for efficient uncertainty 2011). Xiao et al. (2016) perform monotonicity analysis on
analysis (Li and Zhang 2011; Chen et al. 2019b). Active probability transformations of the random variables. This
subspace (AS) emerged as a popular DRM in the last dec- allows capturing the relations between the interval dis-
ade which typically identifies a one-dimensional structure tribution parameters and probability bounds of the struc-
in the map from inputs, irrespective of the dimension of the tural response which is then used to develop parametrized
problem. Constantine et al. (2015) reduce a computation- p-boxes. Liu et al. (2018a) propose an optimized sparse
ally expensive (9500 CPU hours/run) 7 variable problem to grid numerical integration to calculate the bounds of the
a one variable problem. UQ of satellite conceptual design statistical moments of the response function and the cumu-
is performed in Hu et al. (2015). They employ bootstrap lants which are then used with a saddlepoint approxima-
to identify confidence intervals of the AS and verify the tion to obtain the whole range of probability bounds of the
results with an MCS estimate. Often, metamodels are built response function. Simon and Bicking (2017) proposed a
in the AS for analysis purposes (Hu et al. 2017; Jiang and hybrid approach to model and analyze reliability estimates.
Li 2017; Ji et al. 2019). Duong et al. (2019) solve a multi- They use p-box models, acyclic graphs, and belief functions
objective formulation using AS and PCE. Hu et al. (2017) to account for different types and levels of uncertainty infor-
present an alliance algorithm to solve a multidisciplinary mation available. Liu et al. (2018a) use an optimized uni-
optimization where the in-loop UQ is achieved by using AS. variate DRM to compute the bounds of statistical moments
Tripathy et al. (2016) proposed a probabilistic version of AS which is then utilized to identify the p-box from the family
that does not require gradient information and works well of Johnson distribution.
in high dimension. Recent advances include a study that Analytical and sampling approaches are widely used to
combines active learning, active subspace, GP, and MCS solve time-dependent reliability analysis (Hu and Du 2013a,
(Zhou and Peng 2020); a deep learning-based decomposi- b, 2015). Surrogates are also used extensively in reliability
tion of high-dimensional input variables to low-dimensional estimates that require time series data (das Chagas Moura
latent space (Li and Wang 2020); and using it for reliability et al. 2011; Wang and Wang 2012; Kaymaz and McMahon
analysis, a deep learning-based high-dimensional UQ with 2005; Zhang et al. 2017; Wang and Chen 2017; Hawchar
AS (Tripathy and Bilionis 2018). et al. 2018; Wang and Matthies 2019; Wang et al. 2019c).
Wu et al. (2018) propose to use an inverse UQ under the
Bayesian framework which allows capturing the uncertain-
ties in its estimates rather than merely determining the best-
fit values. They project the time series data on to principal

13
 E. Acar et al.

component subspace and perform the propagation in the 4.2.1 Interval‑based approaches

reduced space. Recent developments include learning-based
concepts such as transfer learning for time-dependent reli- Qiu and Wang (2003) present a numerical interval analysis
ability prediction (Zafar and Wang 2020). method for the dynamical response of structures with UBB
external excitation and parameters. Taylor series was used
4.2 Non‑probabilistic techniques to obtain the bounds of the interval dynamical response vec-
tor. The interval structure parameter vector was decomposed
Difficulty in obtaining sufficient data from complex engi- into the sum of the nominal vector and deviation vector from
neering systems leads to the development of non-probabil- which the reliability of the dynamic response structure was
istic methods. Non-probabilistic approaches, unlike proba- calculated. In control theory, Wang et al. (2019b) proposed
bilistic techniques, can operate with small sample size to a closed-loop controller design based on the concept of non-
treat uncertainties. The non-probabilistic approaches can be probabilistic time-variant reliability-based optimization for
attributed to stem from the argument of lack of credibility in structural vibration suppression problems. The boundary
probabilistic analysis results (Annis 2004) when the sample rules of the output response are deduced from the state-space
size is small. The choice of opting for a non-probabilistic transformation, and this transformation is converted to an
approach depends on the degree of extraction of information interval function, and bounds are calculated using Taylor
from the available scarce description, need of non-probabil- expansion. The optimal control parameters are obtained
istic information, computational complexity, and expenses through the particle swarm algorithm. Luo et al. (2021)
(Moens and Vandepitte 2005). The initial studies on non- proposed the conversion of a probability-based reliability
probabilistic approaches were presented by Ben-Haim and optimization problem into a possibility-based reliability
Elishakoff (1995) illustrating the use of non-probabilistic optimization problem where the constraints on the ultimate
convex models for reliability analysis in structural optimiza- bearing capacity are mapped to interval functions using a
tion. Recent years have witnessed a surge in the application satisfaction degree and the optimization is solved using the
of non-probabilistic approaches in the areas of UQ (Gersem method of Simulated Annealing. In line with this idea, Meng
et al. 2006; Wang and Matthies 2019), aerospace engineer- et al. (2019) proposed an active learning method with Krig-
ing (Gersem et al. 2006; Wang and Matthies 2019), (Zheng ing metamodel to rebuild the limit-state function only by
and Qiu 2018), controller design (Wang et al. 2019b), and considering the most concerned point for reliability analy-
structural reliability and analysis (Qiu and Wang 2003; Kang sis. The importance learning function is used to select the
et al. 2011; Hao et al. 2017; Guo and Lu 2015; Meng et al. most concerned point, and the Kriging limit-state function
2016; Jiang et al. 2013a; Meng et al. 2019; Cheng et al. is built around the neighborhood of most concerned point
2020; Luo et al. 2021). and demonstrated using a non-probabilistic approach. When
Non-probabilistic approaches can be broadly classified mixed uncertainties are available (Jiang et al. 2012b), param-
into two categories: (1) interval-based approach and (2) eters with sufficient information were treated as random
convex model-based approach. Recent advances in interval variables while others with less information were treated
arithmetic led to the progress of interval-based non-probabil- as interval variables. When both interval and random vari-
istic approaches. By definition, an interval scalar consists of ables are present (Yoo and Lee 2014), through a sampling-
a single continuous domain in the domain of real numbers ℝ . based approach, the probability of failure can be directly
The range is bounded by a lower and an upper bound. Vari- used to obtain the worst combination of interval variables.
ables in which bounds can be used to represent uncertainties Parameter intervals of the basic variables in conjunction
are modeled using interval analysis approach. Such uncer- with conventional reliability theory (Qiu et al. 2008) allow
tainties are referred to by the term uncertain-but-bounded obtaining system failure probability interval. Based on the
variables. Whereas, in the convex model approach, all the shape function and interval analysis, Liu et al. (2016) pro-
possible values of the uncertain parameters are confined to posed an inverse approach to compute the dynamic load.
a multidimensional convex set. Convex models are preferred Interval process model extends the interval method (Li et al.
in problems where the uncertain parameters are expected 2019a) for quantifying time varying parameters. Instead of
to possess a correlation, as the interval approach assumes using a probability distribution, an interval parameter is used
that the parameters are uncorrelated. These non-probabilistic to characterize the imprecision of the time varying param-
methods are predominantly used in the fields of reliability eter at arbitrary time. Shi and Lu (2019) analyze the safety
analysis and optimization and UQ. The review explores the of dynamic structure by a new dynamic reliability analysis
different non-probabilistic techniques used to qualify the while accounting for both random and interval input param-
safety level of structures with inherent uncertainties. eters by constructing a second-level limit-state function

13
Modeling, analysis, and optimization under uncertainties: a review

4.2.2 Convex model‑based approach 4.3 Remarks on uncertainty analysis

Convex model-based non-probabilistic approaches for reli- Oftentimes, the choice of the approaches is dependent on
ability analysis are used under the circumstances where the computational budget and details available. Based on
the uncertain parameters are correlated and when interval these features, the approaches are compared in Table 1. This
models result in an over-conservative design. Convexity permits the user to choose appropriate approaches based on
approaches employ interval analysis (Moore et al. 2009) the type of data that is available.
and several geometries such as ellipsoid, parallelepiped A table comparing the various techniques under the dif-
(Jiang et  al. 2015), and super ellipsoid (Elishakoff and ferent approaches discussed, based on features such as algo-
Bekel 2013) to bound the uncertain region. Ben-Haim and rithmic complexity, ease of implementation, and computa-
Elishakoff (2013) introduced ellipsoid models to applied tional time is discussed in Table 2, along with the remarks
mechanics field. Ben-Haim (1994) introduced the first non- for the particular technique. Each shaded circle is a score.
probabilistic reliability concept which was based on convex More the score, the method scores well in that feature.
models. Elishakoff and Zingales (2003) studied the differ-
ences between probabilistic and non-probabilistic, anti-
optimization analyses of uncertainty and concluded that for 5 Design optimization under uncertainties
near-unity reliability range these two approaches tend to
each other. Wang et al. (2011) introduced a convex model Uncertainties which may affect performances of engineer-
for structural reliability where ellipse is used to capture the ing systems are ubiquitous in the real world so that design
region of uncertainty. They concluded that both probabilistic optimization under uncertainties has been developed and
approach and convex models are compatible, and the latter exploited in various engineering applications. As shown
can be used where probabilistic information is not available. in Fig. 3, the design optimization under uncertainties is
Ellipse-based convex model and interval analysis approach classified in this paper according to the type of uncertainty
are compared by Qiu and Wang (2005). They observe that that it deals with.
if one knows the form of the convex set, convex models When only aleatory uncertainty exists, RBDO is uti-
should be used, else interval analysis is a more practical lized to find a reliable optimum design. Research on
approach. Super-ellipsoid model for uncertainty analy- RBDO under aleatory uncertainty assumes that there is no
sis was introduced by Elishakoff and Bekel (2013). Jiang epistemic uncertainty, which means that the input statisti-
et al. (2015) introduced a non-probabilistic safety measure cal model is perfectly known and the simulation model is
based on multidimensional parallelepiped model. Here, the accurate. Thus, it mainly focuses on efficient and accurate
uncertainty is characterized by the interval approach with multidimensional integration for reliability analysis and
closed bounds. The works of Kang et al. (2011) illustrate a sensitivity analysis for gradient-based design optimization
multi-ellipsoid convex model-based non-probabilistic reli- since the simulation model is commonly computationally
ability index to treat boundary uncertainties. The proposed expensive. RBDO under aleatory uncertainty is classi-
method only requires implicit forms of limit-state function fied into analytical approaches and sampling approaches
in seeking the concerned performance point for solving the according to how to evaluate the reliability. If a limit-
structural optimization problem. state function is approximated, it mainly utilizes MPP as

Table 1  Remarks on uncertainty analysis approaches

Approaches Sample/time required Information needed for limit- Model multiple failure modes Remarks
state

Sampling More Not required Can be applied directly Simple implementation but time
consuming
Expansion Less Not required Requires algorithm modifica- Quick but need to watch out for
tion approximation errors
Analytical Very less Required (can be approxi- Requires complex algorithm Distribution information
mated) and implementation required
Dimension Reduction Very less Required (can be approxi- Requires complex algorithm Distribution information
mated) and implementation required
Non-probabilistic Less Not Required Can handle In early stages of design,
extremely useful to provide
ball park numbers on proba-
bilistic quantities

13
 E. Acar et al.

Table 2  Table of comparison between various available uncertainty analysis techniques

Modelling Com- Implementation Computational Remarks
plexity Time
Sampling approaches
MCS – Accuracy depends upon the number of
samples
Importance sampling – Requires knowledge of variance of the
distribution sample
– Sampling density choice influences out-
come
Adaptive/Sequence – Requires prior information of the ac-
Sampling quired samples
– Different strategies available for differ-
ent class of problem
Expansion approaches
Taylor series – Requires explicit form of derivatives of
the function
– Prone to errors when higher order terms
are truncated
Perturbation methods – Ability to solve highly nonlinear equa-
tions
– Selection of perturbation variable needs
special attention
PCE – Not very efficient for non-normal ran-
dom input distribution
– Under white noise, performance dimin-
ishes
Analytical approaches
FORM/SORM – When underlying variables are non-
normal, errors are incurred during trans-
formation and handling multiple modes of
failure need modification to the base for-
mulation
– Cannot handle highly nonlinear LSFs
Dimension reduction approaches
Eigen dimension – Does not require sensitivity of system re-
reduction sponses
– Permits statistically correlated and non-
normally distributed random inputs. How-
ever, careful evaluation in case of higher
order correlated terms is necessary
Active subspace – Very attractive for prohibitively expen-
sive computer models.
– Requires gradient information and varia-
tion in the non active subspace is not pre-
served
Non-probabilistic approaches
Interval analysis – Suitable when no information other than
bounds are available but the approxima-
tions are usually conservative or corre-
spond to worst case
– Can deal only with independent uncer-
tain variables
Co nvexity approaches – Suitable when less data but more than
the bounds are available
– Can deal with dependent uncertain vari-
ables
Fuzzy, possibility – Suitable when subjective information
theory about the process needs to be accounted
– Very attractive for exploring designs in
the conceptual design stage

13
Modeling, analysis, and optimization under uncertainties: a review

Design optimization under uncertainties

RBDO under aleatory uncertainty Design optimization under epistemic uncertainty

Analytical approach Sampling approach Input model uncertainty Output model uncertainty

Approximation of Surrogate Non-probabilistic Surrogate model

integration modelling approach uncertainty

Reformulation of Numerical Probabilistic Simulation model

optimization integration approach uncertainty

Fig. 3  Design under uncertainty tree structure

a reference point for the approximation. On the other hand, the uncertainty, which are generally known to be effective
the sampling approach requires repeated simulations for when input data are extremely scarce, and (2) probabilistic
reliability estimation so that a surrogate model is generally approaches to model the input uncertainty through known
utilized to alleviate the significant computational burden. PDF with its epistemic uncertainty existing in distribu-
The analytical approaches can be further categorized into tion parameters and types. The output model uncertainty
various methods depending on (1) how to approximate where the epistemic uncertainty comes from inaccurate
multidimensional integration, and (2) how to reformulate output responses can be also classified into two: (1) sur-
design optimization. The sampling approaches can be clas- rogate model uncertainty caused by the limited number of
sified into two categories: (1) sampling approaches with simulation samples for surrogate modeling, and (2) simu-
surrogate modeling whose main focuses are efficient sur- lation model uncertainty represented as unknown model
rogate modeling and sequential sampling methods, and (2) parameters, discrepancy between experimental results and
sampling approaches with efficient numerical integration simulation response, and measurement error.
for reliability and its sensitivity analysis. In addition to
RBDO under aleatory uncertainty which mainly focuses 5.1 Design optimization under aleatory uncertainty
on component design, there have been approaches which
deal with system design optimization under uncertainties. This section explains RBDO research under aleatory uncer-
Hence, the system RBDO methods are briefly reviewed in tainty where input statistical models are assumed to be
this section as well. given. Thus, the multidimensional integration to estimate
Design optimization under both aleatory and epistemic reliability is the main focus of research. Regardless of
uncertainties has concerned lack of knowledge, data, and approaches, they aim to obtain an accurate optimum satis-
information. Thus, the epistemic uncertainty is classified fying the constraints with the minimum number of computer
into input and output model uncertainty. Since reliabil- simulations. For accuracy comparison, the optimum design
ity is the probability that response of interest is satisfied obtained using MCS is usually used as a benchmark.
under input uncertainty, both input and output uncertain-
ties can affect the reliability analysis and lead to an unreli- 5.1.1 Analytical approaches
able optimum. The input model uncertainty means that an
accurate input statistical model is unknown but insufficient Analytical approaches are traditional ways to approximate
input data are available. Thus, UQ explained in Sect. 3 is the limit-state function. Motivation of the approaches can
required to handle or manage input uncertainties. There be classified into two ways. First, the improvement of inte-
are two approaches to deal with the input model uncer- gration has been developed. The multidimensional integra-
tainty: (1) non-probabilistic approaches such as interval tion can be approximated through moment-based methods
analysis, fuzzy set theory, and evidence theory to describe and MPP-based approximations such as FORM. Besides,

13
 E. Acar et al.

many methods for accurate approximations on multidimen- MPP-based DRM utilizing the history of MPP search simi-
sional integration are proposed. Second, reformulation of lar to approximated SORM, and it shows the same efficiency
double-loop RBDO which requires heavy computations has as FORM while maintaining the accuracy of DRM. Park
been discussed. To this end, decoupled loop and single-loop et al. (2020) developed selective DRM which allocates the
RBDOs are proposed to reduce the computational burden. integration points using a statistical mode selection method
Approximation of integration Approximation of a limit- such as Akaike information criterion.
state function and its multidimensional integration for reli- On the other hand, there have been studies on moment-
ability analysis affect efficiency and accuracy of RBDO. based RBDO which have advantages of not having difficul-
FORM has been widely used and developed in many differ- ties of MPP search and multiple MPP problems in reliability
ent ways based on MPP such as PMA and reliability index analysis (Li and Zhang 2011). Huang and Du (2006) directly
approach as mentioned in Sect. 4 (Hasofer and Lind 1974; estimated cumulative generating function of the response
Tu et al. 1999; Chiralaksanakul and Mahadevan 2005). Lin through moments from the dimension reduction and sad-
et al. (2011) proposed a modified reliability index to prevent dlepoint approximation. Kang and Kwak (2009) exploited
divergence of FORM using a new definition of reliability the maximum entropy principle for PDF modeling and
index since traditional reliability index fails to find the true RBDO. Ju and Lee (2008) combined the moment-based
MPP when the origin in the U-space is within the failure RBDO and surrogate model, and Rajan et al. (2020) also
region. Du and Hu (2012) proposed FORM with truncated developed RBDO using higher-order moments of responses
random variables. Zhang and Du (2010) tried to improve the using local surrogate modeling to overcome the limitations
accuracy of FORM while maintaining a similar level of effi- in MPP-based methods.
ciency. It can be achieved by univariate DRM with quadratic Reformulation of optimization To reduce the compu-
functions and saddlepoint approximation. Recently, Chen tational burden of conventional double-loop RBDO, two
et al. (2019a) integrated SORA with identification of mul- approaches have been studied: decoupled loop and single-
tiple MPPs so that it can be treated in decoupled RBDO. loop RBDOs. In the decoupled loop RBDO, reliability
SORM approximates the limit-state function by a quad- analysis and design optimization are sequentially performed
ratic polynomial function (Breitung 1984). Lee et al. (2012) until convergence. On the other hand, design optimization
and Park and Lee (2018) developed the novel SORM to only is performed in the single-loop RBDO, and reliability
improve the accuracy of SORM through orthogonal transfor- analysis is perfectly approximated without any optimization
mation and integration using general chi-square distribution, loop.
where errors due to approximating the quadratic function For the decoupled loop RBDO, Du and Chen (2004) pro-
by parabolic surface and calculation of reliability are elimi- posed SORA where a double loop is decoupled into reli-
nated. Its sensitivity analysis for RBDO is also developed ability analysis and deterministic optimization using equiva-
by Yoo et al. (2014). Mansour and Olsson (2014) devel- lent deterministic constraints which are shifted to a feasible
oped a closed-form expression for reliability by eliminat- direction according to reliability analysis. Hence, its compu-
ing the rotation of the Hessian matrix. On the other hand, tational efficiency is greatly improved. Zou and Mahadevan
Lim et al. (2014) exploited the symmetric rank-one update (2006) proposed a direct decoupling approach which also
to approximate the Hessian matrix using the path of MPP decoupled the double loop, but reliability analysis is per-
search in approximated SORM to reduce the computations formed through a sampling method instead of MPP-based
for the Hessian matrix calculation. Huang et al. (2018) esti- approximation. Cho and Lee (2011) proposed the improved
mated a cumulative generating function through quadratic SORA where the shifted constraint in the deterministic
approximation and saddlepoint approximation to compute optimization is replaced by a convex linearized function
the reliability analytically. using the gradient and function value obtained from reli-
DRM is developed to approximate the multidimensional ability analysis. Thus, no additional function evaluation is
integration of a limit-state function as a summation of func- necessary, and it is shown that its convergence is better than
tions with reduced dimension. The univariate DRM is most SORA. On the other hand, Chen et al. (2013) proposed adap-
widely used in RBDO which shows more accuracy than tive decoupling approach to enhance the efficiency of SORA
FORM and efficiency than SORM (Rahman and Xu 2004; using novel update angle strategy and feasibility-checking
Lee et al. 2008b). Youn and Xi (2009) proposed EDR to method. The update angle strategy can reduce the necessary
improve the accuracy of DRM by choosing samples along function evaluations, and the feasibility-checking method
the eigenvectors to incorporate the statistical correlation. enables to assess only violated and active constraints. Huang
Kang et al. (2017a) developed the so-called HeDRM to et al. (2016) proposed the incremental shifting vector utiliz-
reduce the effect of cross-terms of univariate DRM through ing the information of previous shifting vector. Chen et al.
rotational transformation. Jung et al. (2019a) further tried (2018) also proposed a probabilistic feasible region approach
to reduce necessary number of function evaluations for for RBDO to further enhance the efficiency of SORA by

13
Modeling, analysis, and optimization under uncertainties: a review

selectively assessing the reliability of each limit-state func- (2017) developed local range RBDO including two phases
tion. Wang et al. (2020) combined the SORA with moment which are to find the local range based on SVM and con-
method through univariate DRM and PDF estimation such struct an accurate Kriging model in the local range. Li et al.
as maximum entropy method. (2019c) developed a sequential surrogate model method for
On the other hand, the single-loop approach has been RBDO using RBF by sequentially refining the surrogate
also exploited widely due to its high efficiency compared model in the vicinity of the current design. PCE has been
to conventional RBDOs. Various single-loop RBDOs to also popularly investigated to be used in RBDO. Hu and
eliminate the reliability analysis such as single-loop single Youn (2011) proposed the adaptive-sparse PCE to overcome
vector (Chen et al. 1997; Yang and Gu 2004) and single- the curse of dimensionality by automatically detecting sig-
loop approach (Liang et  al. 2008) have been proposed, nificant polynomials and adjusting the PCE order. Zhou and
where the Karush–Kuhn–Tucker optimality condition is Lu (2019) developed the Bayesian compressed sensing for
used to replace probabilistic constraints (Mohsine et al. PCE surrogate model and integrated it with the active learn-
2006; Agarwal et al. 2007). Shan and Wang (2008) proposed ing strategy for RBDO. Zhu and Du (2016) proposed the
reliable design space to further improve the efficiency of dependent Kriging prediction to consider the correlations
single-loop approach by approximating the gradient vector between prediction at realizations of input random vari-
at MPP. Jeong and Park (2017) proposed single-loop single ables for MCS, so that a new learning function accounting
vector using the conjugate gradient where the convergence for variation of reliability is used for sequential sampling.
of single-loop single vector is improved by MPP estima- Wang and Wang (2014) developed cumulative confidence
tion through conjugate gradient. Jiang et al. (2017) proposed level to quantify the accuracy of reliability estimation using
adaptive hybrid single-loop method that is more suitable for a surrogate model, and then a sequential sampling approach
non-linear RBDO sacrificing little efficiency by checking for RBDO is adopted based on cumulative confidence level.
the feasibility of the approximate MPP. Similarly, Li et al. Numerical integration methods Au and Beck (2001) and
(2019d) proposed a new oscillating judgment criterion and Au (2005) developed the subset simulation for reliability
adaptive modified chaos control method to select the MPP analysis and its sensitivity analysis, and Li and Cao (2016)
search formula and integrate with single-loop approach. developed Matlab code for the subset simulation and struc-
tural optimization. Lee et al. (2011) proposed stochastic sen-
5.1.2 Sampling approaches sitivity analysis of reliability and statistical moments when
the input random variables are correlated. The score func-
Sampling approaches are divided in this section into two: tion is used to derive sensitivities and the input statistical
strategies for more accurate and efficient surrogate modeling model is described with parametric marginal distribution
during RBDO such as a sequential sampling, and various and copula function. Cho et al. (2016b) developed the sam-
numerical integration methods. The sequential sampling pling-based RBDO when standard deviations vary. Dubourg
attempts to find the best sampling point to be added in the et al. (2013) proposed the metamodel-based IS using quasi-
current sample set to best improve accuracy of a surrogate optimal instrumental PDF using probabilistic classification
model. The numerical integration method attempts to com- functions defined by a Kriging model. Zhu et al. (2015) pro-
pute the reliability during optimization with less number of posed a new sampling-based RBDO via score function with
samples than MCS. a reweighting scheme so that its computational efficiency
Strategies for accurate and efficient surrogate modeling- is improved. Recently, Chaudhuri et al. (2020) proposed IS
Dubourg et al. (2011) built a Kriging model on augmented reusing information accumulated from past iterations, which
reliability space and a strategy to sequentially refine the is exploited in RBDO to reduce the computational burden.
Kriging model, where the reliability is estimated through
subset simulation. Bichon et  al. (2008, 2013) integrate 5.1.3 System RBDO
efficient global reliability analysis into RBDO. Chen et al.
(2014, 2015) proposed a local adaptive sampling for RBDO Component RBDOs discussed so far find an optimum where
to improve the efficiency of constructing a Kriging model multiple constraints are independently satisfied, whereas
based on constraint boundary sampling (Lee and Jung 2008), system RBDO finds an optimum with a single system con-
and important boundary sampling accounting for objective straint aggregating all probabilistic constraints. Thus, an
function, which is integrated with SORA. Similarly, Meng effective approach to obtain the system reliability is the
et al. (2018) exploited the adaptive directional boundary key idea of system RBDO studies. Ba-Abbad et al. (2006)
sampling accounting for objective function. Li and Cao proposed a modified approach for RBDO of series systems
(2016) proposed a local approximation method using MPP adapting SORA where the optimizer distributes the reliabil-
to check the feasibility of constraints and locally refine the ity of the system over its components while constraining
surrogate model around MPP. On the other hand, Liu et al. the system reliability only. Liang et al. (2007) proposed a

13
 E. Acar et al.

single-loop system RBDO that enables to distribute each tar- instance, the simple Taylor expansion (Lee and Park 2001;
get reliability of failure mode, so that only system reliability McAllister and Simpson 2003) and perturbation methods
has to be assigned. McDonald and Mahadevan (2008) also (Doltsinis and Kang 2004) are used for the moment esti-
developed the equivalent formulation including both compo- mation, Xu and Rahman (2004) proposed the univariate
nent and system reliability constraints as single-loop RBDO. DRM for multidimensional integration, and Youn et al.
Song and Kang (2009) and Nguyen et al. (2010) proposed (2005b) proposed the performance moment integration.
matrix-based system reliability method and system RBDO In practical, reliability-based robust design optimization
using matrix-based system reliability, so that it can account (RBRDO) rather than RDO is much effective in consider-
for statistical dependence between component events. Lee ing two objectives, which are reliability of constraints and
et al. (2010) exploited the MPP-based DRM to calculate robustness of product quality (Lee et al. 2008a; Youn and
the component reliability and integrate with Ditlevsen’s sec- Xi 2009; Motta and Afonso 2016). The multi-objective
ond-order upper bound according to convexity of limit-state cost function is one of the key issues to treat combined
functions. Wang and Wang (2015b) developed the integrated robustness measures. Yadav et al. (2010) proposed a multi-
PMA exploiting GP model to accurately estimate system objective framework by addressing various quality losses
reliability. Xiao et al. (2020) developed a system active simultaneously, Sun et al. (2011) showed multi-objective
learning Kriging for system RBDO based on expiration risk RDO on crashworthiness design of the vehicle to gener-
function to refine the Kriging model iteratively. ate Pareto solutions, and Shahraki and Noorossana (2014)
presented RBRDO using an evolutionary multi-objective
5.1.4 Robust design optimization (RDO) genetic algorithm. Meanwhile, epistemic uncertainty
induced by lack of knowledge also has been treated in
Robust design optimization (RDO) attempts to improve RDO. Tang et al. (2012) developed RBRDO accounting
the product quality by minimizing the variability of an for both reliability and robustness indices under epis-
output response propagated from input uncertainty. In temic uncertainty represented by info-gap theory. Kang
other words, the optimal design gives a high degree of and Bai (2013) proposed a new robustness measure and
robustness that is relatively insensitive to input uncertain- RDO based on a convex model for uncertain-but-bounded
ties, and several studies have widely investigated RDO parameters. Zaman and Mahadevan (2013) developed
frameworks to systematically organize the existing stud- RDO for multidisciplinary systems accounting for both
ies (Zang et al. 2005; Beyer and Sendhoff 2007; Schuëller aleatory and epistemic uncertainties, and Li et al. (2020)
and Jensen 2008; Chatterjee et  al. 2019). Specifically, recently treated both parameter and model uncertainty in
RDO concentrates on estimating the first two statistical multidisciplinary RDO. In engineering application, Ghisu
moments of the output response and their sensitivities et al. (2011) presented the RDO of the gas turbine sys-
(Sandgren and Cameron 2002). Thus, there exist concep- tem, and Fang et al. (2015) performed RDO for fatigue
tual differences between RBDO and RDO. RBDO usually life to design truck cap, and Lee et al. (2020b) exploited
treats constraints of catastrophic failure in rare extreme RDO to the thermoelectric generator system. Especially in
events where the cost function to be minimized is gener- the composite structure, das Neves Carneiro and António
ally deterministic, while RDO emphasizes the response (2019) presented the RBRDO of angle-ply composite lam-
sensitivity with respect to the input variations or allows inate structure accounting for both weight and determinant
for the maximum possible system variability. The efficient of covariance of response. Zhou et al. (2019a) exploited
and accurate estimation of the statistical moments is the RDO to variable angle tow composite structures under the
key to RDO research. Numerical integration such as MCS material and applied load uncertainties, and several studies
is the most intuitive way but requires a heavy computa- on RDO of composite structures can be found (António
tional burden, and thus the surrogate model to replace the and Hoffbauer 2009; Bacarreza et al. 2015).
simulation model can resolve it. Chakraborty et al. (2017)
proposed RDO using polynomial correlated function
expansion called high-dimensional model representation. 5.2 Design optimization under both aleatory
Coppitters et al. (2019) used PCE to emulate the physical and epistemic uncertainties
model for RDO of the photovoltaic–electrolyzer system.
Recently, Chatterjee et al. (2019) presented an extensive All RBDO studies introduced in the previous section
survey to illustrate the performance of surrogate models in assume that statistical models of input random variables
RDO, Keane and Voutchkov (2020) proposed a combined are given, and a simulation model gives exact responses.
Co-kriging model for RDO. On the other hand, many However, it is very difficult to know the statistical mod-
approximation methods on integration have been pro- els of all input random variables in practical engineer-
posed to reduce the number of function evaluations. For ing applications, and the simulation model always shows

13
Modeling, analysis, and optimization under uncertainties: a review

discrepancy compared with the experimental results. Thus, parameters. The non-probabilistic reliability index based
estimation of the input statistical model needs to be per- on a multi-ellipsoid convex model is used to quantify
formed using collected data as discussed in Sect. 3, and the reliability in RBDO. Guo and Lu (2015) presented
the estimated statistical model could be inaccurate when a mixed interval-convex model non-probabilistic RBDO
insufficient data are used. Also, inaccuracy of a surrogate methodology for structures with uncertain-but-bounded
model emulating the simulation model can be epistemic parameters, where the input uncertain parameters were
uncertainty in RBDO. In this section, studies to find an treated as interval variables. Meng et al. (2016) developed
RBDO optimum accounting for epistemic uncertainties a decouple approach for the non-probabilistic RBDO by
such as input model uncertainty and output model uncer- shifting constraints of deterministic optimization, and Hao
tainty including surrogate and simulation model uncer- et al. (2017) proposed an efficient adaptive-loop method
tainty are discussed. for the non-probabilistic RBDO aiming at improving effi-
ciency, where the convex model is used to describe the
5.2.1 Input model uncertainty input uncertainty. Recently, Kanno (2019) developed a
data-driven non-parametric RBDO accounting for confi-
Non-probabilistic approach This section introduces vari- dence level of reliability.
ous design optimizations where input model uncertainty Probabilistic approach This section introduces develop-
is described in non-probabilistic ways. Interval analysis ments in RBDO where input statistical model is described
is carried out when only nominal value and its lower and by a probabilistic way such as the Bayes’ theorem where
upper bounds are available through scarce data or engi- the input distribution parameters can be adjusted or follow
neer’s experience. Du and Chen (2004) and Du et  al. a posterior distribution obtained from input data. Youn and
(2005) proposed a design optimization with the mixture Wang (2008) proposed a Bayesian RBDO combined with
of random and interval variables. Since identification of EDR when the input statistical model is unknown. Noh
the worst case interval variables requires heavy computa- et al. (2011a, b) developed RBDO with confidence level by
tions, the sequential single loop is employed to improve adjusting input standard deviations and correlation coeffi-
the efficiency. Yoo and Lee (2014) developed sampling- cients when input data are not sufficient. Cho et al. (2016a)
based design optimization in the presence of interval proposed a conservative RBDO using conservativeness of
variables. Cho et  al. (2020) developed an MPP search reliability where input distribution parameters and types
method for mixture of random and interval variables and are quantified through the Bayesian approach. Following
its sensitivity analysis where both types of variables can the previous frameworks, Jung et al. (2019b, 2020a) tried to
be iteratively updated to find a correct MPP. Fuzzy vari- reduce the computational cost for the conservative RBDO
ables described using membership functions are employed exploiting the MPP approach in the space of distribution
in possibility-based design optimization. Du et al. (2006) parameters and developed the bi-objective confidence-based
integrated PMA with possibility-based design optimiza- design optimization to determine the optimal number of
tion improving the efficiency of the optimization by using input data. Zaman and Mahadevan (2017) exploited a four-
the maximal possibility search method. Youn et al. (2007) parameter flexible Johnson family of distribution to describe
considered product quality loss analogous with robustness the input statistical model. Moon et al. (2018) extend the
of quality and integrated it with possibility-based design previous approach to biased simulation models so that both
optimization. Lee et al. (2013) compared possibility-based input and output test data are used to estimate the distribu-
design optimization with RBDO using confidence level tion of reliability. Ito et al. (2018) proposed the conservative
and measured conservativeness of an optimum for each reliability index that can be decomposed into target reliabil-
approach through mathematical and engineering examples. ity index and epistemic reliability index, so that aleatory
The evidence theory uses two measures which are belief and epistemic uncertainties of input random variables can
and plausibility to quantify the bounds of the precise prob- be considered simultaneously. Moon et al. (2019) exploited
ability. Mourelatos and Zhou (2006) used the evidence a bootstrapping method for bandwidth to obtain the distribu-
theory to assess reliability with incomplete information, tion of reliability using KDE as the input statistical model.
and a computationally efficient optimization is proposed.
Alyanak et al. (2008) developed a design optimization 5.2.2 Output model uncertainty
method using the evidence theory based on a gradient pro-
jection technique. Jiang et al. (2013b) developed the most The output model uncertainty means that output responses
probable focal element corresponding to MPP in RBDO, of interest would be inaccurate due to insufficient simulation
and then reliability can be obtained efficiently through samples for surrogate modeling and fundamental inability
most probable focal element. Kang et al. (2011) proposed a of a simulation model to numerically emulate the real physi-
non-probabilistic RBDO exhibiting uncertain-but-bounded cal model. Thus, various researches can be included in this

13
 E. Acar et al.

section such as model calibration and validation induced various statistical model calibration and validation strate-
by biased simulation models with unknown parameters and gies, and categorized it for clarity.
quantification of surrogate model uncertainty.
Surrogate model uncertainty Picheny et al. (2008) pro- 5.3 Remarks (Discussion, Consideration) on design
posed margin to the response predicted by a surrogate model and optimization under uncertainties
using biased fitting. Viana et al. (2010) exploited the cross-
validation method to determine safety margin of a surro- In this section, several remarks on design optimization
gate model. Zhao et al. (2013) developed weighted Kriging under uncertainties are given to arrange pros and cons of
variance for sampling-based RBDO using Akaike informa- each approach, potential uses, and promising perspectives.
tion criterion. An and Choi (2012) proposed the Bayesian Design optimization under aleatory uncertainty mainly treats
framework incorporating the input model and surrogate a multidimensional integration on joint PDF of random vari-
model uncertainties quantified as hyper-parameters of Krig- ables in this paper. The analytical approach focuses on the
ing to perform integrated reliability analysis. The surrogate approximation of a limit-state function at MPP, so that it is
model uncertainty in reliability analysis quantified as cor- very efficient but cannot ensure accuracy since non-linearity
related prediction in Kriging is also taken into account in the of the limit-state function is unknown. In particular, it can be
research of Nannapaneni et al. (2016). Li and Wang (2018) much erroneous when decoupled and single loops requiring
proposed confidence-driven design optimization to avoid additional approximations are used instead of double loop.
underestimation of reliability in RBDO due to insufficient The alternatives such as the moment-based method can alle-
simulation data. Li and Wang (2019) also developed RBDO viate the difficulty induced by approximation at MPP, but it
accounting for surrogate model UQ using equivalent reliabil- still requires estimation of higher moments and the approxi-
ity index exploiting the Gaussian mixture model. Jung et al. mation of parametric PDF. Thus, it can be addressed that an
(2021) recently proposed the confidence-based design opti- analytical approach based on approximation is suitable for
mization accounting for distribution of reliability induced a high-dimensional problem or a problem where simulation
by surrogate model uncertainty. is extremely expensive so that surrogate modeling is not
Simulation model uncertainty Simulation models such available.
as finite element analysis numerically solve a partial dif- On the other hand, the sampling approach is a more prac-
ferential equation under various assumptions yielding tical method in the real world since it is more reliable than
discrepancy with experimental results. The goal of model the analytical approach, and its convergence is guaranteed
calibration of simulation model is to accurately emulate the as the number of samples increases. In fact, the accuracy of
experiment by estimating unknown model parameters and the sampling approach is highly dependent on the accuracy
model discrepancy. Xiong et al. (2009) proposed the maxi- of the surrogate model since the direct numerical integra-
mum likelihood estimation-based approach to estimate the tion using the simulation model is extremely difficult. Each
distribution parameters of unknown parameters instead of surrogate model has different characteristics, and it is very
the Bayesian approach. Arendt et al. (2012) proposed an challenging to determine which surrogate model is suitable
overall framework for model updating, a modular Bayes- in such a situation. To resolve the difficulty of DoE in sur-
ian approach, so that GP for the simulation model and bias rogate modeling, various strategies to update the surrogate
function and posterior distribution of calibration param- model, also called active learning, have been widely investi-
eters can be obtained. Jiang et al. (2013b) proposed RBDO gated for RBDO. Especially, the method of effectively com-
under model and parameter uncertainties using GP mod- bining active learning and numerical integration could be a
eling exploiting the multi-fidelity structure. Pan et al. (2016) good option in RBDO. Thus, it is addressed that if surrogate
developed a copula-based approach for bias modeling and modeling is available, active learning Kriging combining
unknown parameter calibration, and then model bias can be efficient integration can be recommended since its accuracy
expressed as conditional PDF. Shi and Lin (2016) showed a can be quantified and improved through additional simula-
new RBDO exploiting adaptive response surface using the tions. Epistemic uncertainty induced by insufficient data,
Bayesian metric to prevent inaccurate response surface and knowledge, and information always exists in the real word,
GP for bias modeling. Moon et al. (2017) proposed RBDO but conventional RBDOs assumed that prior information
using confidence-based model validation, which means that and the sufficient number of data are available to estimate
the distribution of reliability is taken into account using all models, and thus it has focused on efficient and accu-
adaptive KDE of insufficient experimental data. Xi (2019) rate numerical integrations for reliability analysis, surrogate
developed confidence-based reliability analysis consider- modeling, and simulation model calibrations. However, we
ing the epistemic uncertainty induced by simulation model addressed that epistemic uncertainty in each process has to
uncertainty for three representative scenarios. Recently, Lee be properly quantified and taken into account in reliability
et al. (2019a) and Jiang et al. (2020) investigated studies on analysis and design optimization for practical uses.

13
Modeling, analysis, and optimization under uncertainties: a review

Design optimization under input model uncertainty can model structure and its parameters, and the output data.
be categorized into non-probabilistic and probabilistic meth- Hu et al. (2017) discussed uncertainty quantification and
ods. Although it depends on the characteristics of the input management in additive manufacturing, and presented cur-
uncertainty, non-probabilistic methods such as the convex rent status, needs, and opportunities. Håkansson (2019)
model, which has been vastly studied recently, are known to used a general uncertainty management framework to ana-
be effective when the input data are extremely limited and lyze the standard uncertainty resulting in the heat transfer
bounded. On the other hand, the probabilistic method using coefficients obtained with sensor experiments. Kumar et al.
joint PDF and Bayes’ theorem can be erroneous when uncer- (2020) demonstrated an efficient uncertainty quantification
tainty quantification of epistemic uncertainty is inaccurate and management in the early-stage design of composite
due to a very small number of data and wrong assumptions applications. Son et al. (2020) addressed statistical model
on PDF, but it can show the convergence as more data are improvement consisting of model calibration, validation, and
available and has a perfectly theoretical background. To the refinement techniques using a case study of an automobile
best of the authors’ knowledge, however, no studies have steering column and then applied the uncertainty modeling
yet to thoroughly compare the effectiveness of these two results to RBDO.
approaches quantitatively or theoretically according to the
number of given data. Design optimization under output 6.1.2 Applications of uncertainty analysis
model uncertainty, such as surrogate model uncertainty
and simulation model uncertainty, has been limited even Cadini et al. (2015) demonstrate sampling-based approach
though it frequently occurs in the real world. Specifically, in radioactive waste repository and Radaideh and Kozlowski
efforts have been made to improve the surrogate model and (2020) provide a nuclear reactor example. Booker et al.
calibrate the simulation model to estimate unknown model (1999) and Hosder et al. (2001) discuss applications in heli-
parameters and model bias. However, epistemic uncertainty copter rotor blade and high-speed civil aircraft, respectively.
cannot be perfectly reduced, and the available data are Iooss and Le Gratiet (2019) develop a functional risk curve
always limited. Therefore, it should be taken into account to in non-destructive testing in aeronautics and steam genera-
cope with various situations in the real world. Unfortunately, tor tubes application. A Krylov space dimension reduction
the studies on surrogate modeling and simulation model approach is demonstrated on a 64-pin radio frequency-inte-
calibration are organized in the literature, but the quantifi- grated circuit-linear time-invariant system by Freund (2003).
cation of output model uncertainty and conservative design A high-speed delft catamaran example is demonstrated on
optimization accounting for it have been limited and should using RBF in Volpi et al. (2015). Bichon et al. (2008) dem-
be developed in the future. onstrate efficient global optimization on bistable MEMS.
Taflanidis and Beck (2008a, 2008b), Jensen et al. (2009),
and Li et al. (2016) discuss examples on x-storey structures
6 Applications, software, and benchmark subject to different types of base loads.
problems In the functional expansion approaches, Ishigami, Sobol,
and Morris functions are discussed in Blatman and Sudret
6.1 Applications (2010a, 2010b) and Zhou and Lu (2019). An urban drainage
simulation is demonstrated in Nagel et al. (2020). Dodson
Though plenty of methods were introduced and developed and Parks (2015) and Wang and Matthies (2019) present
to treat uncertainties, only in the last decade, the methods approaches to perform airfoil-shaped optimization while
were liberally used in industrial applications across different robust design of airfoil is also discussed (Dodson and Parks
domains. In the following, we list the various applications 2015). Wu et al. (2017) discuss a nuclear reactor system
addressed under the different classes. design and Kumar et al. (2020) present a rotor blade design
and associated dynamic studies. Chan et al. (2007) and Du
6.1.1 Applications of uncertainty modeling and Chen (2004) use analytical approaches to propagate
uncertainties in Passive vehicle suspension design and a
Wunsch et al. (2015) presented quantification of combined vehicle crashworthiness of side impact, respectively. Muk-
operational and geometrical uncertainties in turbo-machin- hopadhyay et al. (2016) discuss uncertainty propagation in
ery design. Allen et al. (2015) discussed uncertainty man- composite structures. When the dimensions are large, dimen-
agement in the integrated realization of engineered materials sion reduction techniques are employed to reduce the dimen-
and components. Azevedo et al. (2015) presented the cali- sion in which the propagation or analysis can be performed.
bration of traffic microscopic simulation models for safety Several applications in automotives such as vehicle side
analysis analyzed considering four different key uncertainty impact (Lee et al. 2008a), lower control arm of a vehicular
sources: the input data, the calibration methodology, the system (Youn and Wang 2008), and aerospace applications

13
 E. Acar et al.

such as conceptual sizing (Bae and Alyanak 2016), rotating is also popularly used (Youn and Choi 2004; Youn et al.
disk (Park et al. 2020), hypersonic scramjet (Constantine 2005a). RBDO for crashworthiness of vehicle side impact
et al. 2015), and satellite design (Hu et al. 2015, 2017). Non- has nine design variables and two design parameters, and
probabilistic approaches are used in design of composite the limit-state functions are approximated as polynomials
laminated panels in supersonic flow (Zheng and Qiu 2018) using response surface method, so that it is proper to verify
and ultra-precision high-speed press (Cheng et al. 2020). RBDO as a problem with moderately high dimension (Youn
Interval approaches are used to demonstrate an 8 degrees et al. 2004). On the other hand, the ten-bar truss example has
of freedom vehicle vibration problem (Li et al. 2019a) and maximum stress constraints with knockdown factor, and the
design of an automobile front axle (Shi and Lu 2019). two-story steel frame example has ultimate and serviceabil-
ity limit-states involving bending moment, axial force, and
6.1.3 Applications of RBDO displacement. These examples are also widely recommended
to demonstrate the RBDO for a simple engineering applica-
Allen and Maute (2004) allowed the design of aeroelastic tion framework since it has a quite large number of random
structures where the reliability of structural and aerody- variables, multiple limit-state functions, engineering aspects
namic criteria should be satisfied, and thus the optimized related to simple static analysis, and it is easy to implement.
structure yields significantly improved results. In addition, Hosseinzadeh et al. (2018) and Jo et al. (2021) exploited
there have been many efforts on aerodynamic field apply- the bracket’s FEA model for deflection and modal analy-
ing RBDO framework (Paiva et al. 2014; Nikbay and Kuru sis using Matlab partial differential equation toolbox. Jung
2013). Sun et al. (2017) applied multi-objective RBDO to et al. (2020b) solved the heat transfer problem for the cool-
tailor rolled blank hat-shaped structure, one of the key light- ing flange using COMSOL to calculate the cooling power.
weight technology for vehicle crashworthiness. Youn et al. Park et al. (2020) and Jung et al. (2021) also use the FEA
(2004) and Acar and Solanki (2009) also utilized RBDO in model for crank arm optimization in Hyperstudy, analyzing
vehicle design for crashworthiness. Grujicic et al. (2010) the maximum von-Mises stress under static loading.
exploited RBDO for durability of suspension in the presence
of the uncertainty of material properties and shape param- 6.2 Software
eters. Lee et al. (2019b, 2020a) applied RBDO to electric
vehicle design in reliability-based design for market systems Several commercial software packages provide capabili-
and to shared autonomous electric vehicle design and opera- ties to treat uncertainty. In the following, we provide some
tion under uncertainties. Fan et al. (2019) used the Krig- insights into few software. (All material are from the respec-
ing model to perform RBDO of crane bridges. Hassan and tive software’s website. These are by no means comprehen-
Crossley (2008) proposed RBDO of spacecraft, and Shin sive in terms of the list of software and the capabilities in the
and Lee (2014) exploited RBDO to determine the optimal listed software). Altair offers uncertainty analysis capabil-
radius and speed limit in windy environments under uncer- ity through RAMDO (2021) which includes generation of
tainty. RBDO framework for composite structure has also random numbers to simulate input variability and perform
been widely investigated. Gomes et al. (2011) and Lopez RBDO and RDO. It also has inbuilt dynamic kriging and
et al. (2011) proposed the RBDO of laminated composites variance window approach. Ansys, in its OptiSLang (2021)
structure, and Sohouli et al. (2018) and Duan et al. (2020) product, has reduced order modeling, Robust design, and
exploited the efficient RBDO for a composite structure such Model calibration modules. Dassault Systemes Simulia, in
as decoupled and single loops. Li et al. (2017) developed its Isight (2021) product, has capabilities on different meta-
the framework for RBDO of wind turbine drivetrains under models, MCS with a variety of distributions, Mean value
wind and manufacturing uncertainties. Azarkish and Rashki method, FORM/SORM, and Taguchi techniques for six
(2019) performed reliability analysis and sensitivity analysis sigma robustness. Esteco, in its modeFRONTIER (2021)
of shell and tube heat exchangers. Makhloufi et al. (2016) module, can generate samples from a variety of probabilis-
performed RBDO or wire bonding in power microelectronic tic distributions including capabilities such as MCS, PCE,
devices. Li et al. (2019b) introduced the multidisciplinary Adaptive PCE (Least Angle Regression method is used
RBDO of a cooling turbine blade along with heat transfer to find, rank, and reduce the most significant polynomial
analysis and strength analysis. Ronold and Larsen (2000), terms), and Sobol indices for sensitivity studies. In addi-
Toft and Sørensen (2011), and Hu et al. (2016) conducted tion, frameworks for RDO, RBDO, and Taguchi techniques
RBDO of wind turbine blades. are available. OmniQuest (2021), as part of their Iliad mod-
A two-dimensional mathematical example is one of the ule, offers capabilities in RBDO, RDO, n-sigma Design,
most widely used mathematical formulations in RBDO since MCS, approximate MCS, and approximate Latin hypercube
it can be visualized and has narrow safe region, and modi- sampling on a variety of distributions. SmartUQ (2021)
fied formulation with much non-linear second constraint has capabilities on statistical calibration, quick emulator

13
Modeling, analysis, and optimization under uncertainties: a review

building with large data, propagation of uncertainties using 6.3 Benchmark problems

metamodels, or PCE. Softwares such as Mathematica and
Matlab permit programming all the techniques discussed This section covers several widely used example problems
in this manuscript. UQLab (Marelli and Sudret 2014) is used in structural reliability estimation and RBDO. When-
an open source (academic use) Matlab add on. Several ever a new reliability estimation or RBDO method was
open source packages in python such as UQpy (Olivier proposed, these example problems were used to show that
and Shields 2020) and UQ-Pyl (Wang et al. 2016) are also the proposed method was capable of handling problems
available. A list of such software packages is provided in with the following properties: (1) large number of ran-
UQWorld (2021). dom variables, (2) non-linear limit-state function(s), (3)
noisy limit-state function(s), (4) multiple failure regions

Table 3  Benchmark problems for RBDO

Problem Reference paper(s) Number of random variables Used probability distributions Number of
limit-state
functions

Burst margin of a disk Wang et al. (2004) 6 Weibull, Normal, Uniform 1

Duffing type oscillator Schueller and Pradlwarter 4000 All normal 1
(2007)
Embankment dam Schueller and Pradlwarter 2160 Normal, Lognormal 1
(2007)
Hyperbola with multiple MPPs Engelund and Rackwitz (1993) 2 All normal 1
I-beam problem Wang et al. (2004) 10 All normal 1
N-dimensional hyperplane Engelund and Rackwitz (1993) 2, 10, 50 All normal 1
Noisy limit-state Liu and Der Kiureghian (1991) 6 All lognormal 1
Parallel system Melchers (1989) 5 All normal 4
Shear beam model Schueller and Pradlwarter 225 All normal 1
(2007)
Series system Melchers (1989) 6 All normal 3
Tuned mass damper Chen et al. (1999) 2 All normal 1
Limit-states: Linear, Non-linear, Papaioannou et al. (2015) 20, 100, 100, 200 Standard Normal, Normal 1, 2, 1, 1
Quadratic, Numerical
Non-linear oscillator Echard et al. (2013) 6 Normal 1

Table 4  Benchmark problems for structural reliability estimation

Problem Reference paper(s) Number of random/ Used probability distributions Number of
design variables limit-state
functions

Cantilever beam Wu et al. (2001) 4/2 All normal 2

Composite panel Qu et al. (2003) 12/4 All normal 1
Mathematical example-1 Youn and Choi (2004) 2/2 Normal, Gumbel 3
Mathematical example-2 Aoues and Chateauneuf (2010) 2/2 All normal 1
Short column design Aoues and Chateauneuf (2010) 6/2 Gumbel, Weibull, Lognormal 1
Ten-bar truss Kumar et al. (2009) 27/10 Uniform, Extreme type I, Lognormal 1
Torque arm problem-1 Kim et al. (2006) 8/8 All normal 1
Torque arm problem-2 Acar (2016) 10/7 Normal, Lognormal 1
Two-bar truss Ramakrishnan and Rao (1996) 5/2 Normal, Beta, Gumbel, Lognormal 2
Two story steel frame Aoues and Chateauneuf (2010) 6/2 Normal, Lognormal 5
Vehicle side crash Youn and Choi (2004) 11/9 All normal 6
Fortini’s clutch Lee and Chen (2009) 4 Beta, Normal, Rayleigh 1
Piecewise linear function Breitung (2019) 2 Standard Normal 2,1
Metaball function

13
 E. Acar et al.

(multiple MPPs), (5) multiple failure modes, and (6) vari- ferent approaches are used for modeling convex regions,
ables having different probability distributions. Tables 3 choice of convex models for data without prior infor-
and 4 provide information about the benchmark prob- mation on their characteristics is not available. Hence,
lems used in structural reliability estimation and RBDO, an ensemble of geometries and respective error meas-
respectively. ures along with a continuation to uncertainty design is a
potential research area.
– Both aleatory and epistemic uncertainties are handled in
7 Concluding remarks RBDO to find an optimum design to satisfy a given tar-
get reliability or confidence. Various methods reviewed
Despite the fact that design optimization of structural and in uncertainty modeling and analysis are combined with
multidisciplinary systems under uncertainties has been a RBDO methods under various uncertainty situations.
topic of growing interest over the past decades, the current Research in RBDO mainly focuses on how to improve
literature lacks a thorough review of uncertainty treatment its accuracy or efficiency or both.
practices including uncertainty modeling, uncertainty analy- – RBDO under aleatory uncertainty is classified into MPP-
sis, and design under uncertainty. This article provides a based approach and sampling-based approach. The MPP-
comprehensive review of the uncertainty treatment practices, based approach can be further classified into various
and complement existing reviews on similar subjects (e.g., methods depending on MPP search methods, optimiza-
reliability analysis, reliability-based optimization, uncer- tion formulation, and limit-state function approximation.
tainty representation, sensitivity analysis under uncertainty, The sampling-based approach is usually combined with
and uncertainty handling theories). From this review, we can surrogate models to alleviate its computational burden.
draw the following conclusions and provide recommenda- To further improve computational efficiency of surrogate
tions for future studies: modeling, efficient surrogate modeling combined with
sequential sampling strategies have been studied.
– Probability theory is the oldest and still the most widely – Both input statistical model and output model uncertain-
used uncertainty handling theory, whereas the com- ties are considered in RBDO under epistemic uncertainty.
bined use of various uncertainty handling theories can Its main strategy is to guarantee a reliable or confident
take advantage of the ability of each theory for better optimum design even when insufficient information
uncertainty modeling. Hybrid approaches that focus on is available to model the input or output. In addition
combined use of various uncertainty handling theories is to active research topics that this article covers, vari-
an active area of research. ous machine learning techniques are currently being
– In probabilistic handling, uncertainty is characterized by attempted to further improve accuracy and efficiency of
using a probability distribution function when a paramet- RBDO.
ric approach is used, whereas it is characterized by KDE
when a non-parametric approach is used. Multivariate
modeling that considers correlation between uncertain Acknowledgements  The authors dedicate this paper to Prof. Raphael
T. Haftka, who worked extensively on topics related to uncertainties for
variables is an active area of research. over 3 decades leading to more than 100 contributions in applications
– When moderate samples are available, uncertainty mod- spanning from structural composites to turbomachines and material
eling and propagation are typically through metamodels models. He was a prolific collaborator and worked with numerous col-
and probabilistic approaches. Though metamodel con- leagues from other universities and countries. In that regard, this paper
reflects such an effort with collaborators from three different countries/
struction and usage is widely adopted, models such as universities.
polynomial response surface, Kriging, SVR, and RBF are
mostly used. Usage of stochastic approaches and machine Author contributions  EA—Entire manuscript curation, writing Uncer-
learning algorithms are to be tested and adopted. Propa- tainty modeling, review & editing entire manuscript, resources; GB—
gation of uncertainties through metamodels is an active Writing, review & editing Uncertainty modeling; YJ—Writing, review
& editing Design optimization under uncertainties; IL—Entire manu-
area with potential avenues of improvement in using script curation, writing Design optimization under uncertainties, review
multi-fidelity metamodels for propagation. & editing entire manuscript, resources; PR—Entire manuscript cura-
– In the event of large dimensions, dimension reduction tion, writing Uncertainty analysis, review & editing entire manuscript,
techniques are widely used. However, propagation of resources; SSR—Writing, review & editing Uncertainty analysis.
uncertainties in the reduced dimension and mapping the
errors in the reduced space to the original space along Declarations 
with handling mixed variables are open research area.
Conflict of interest  The authors declare that they have no conflict of
– When available samples are scarce, approaches based on interest.
interval theory and convex models are used. Though dif-

13
Modeling, analysis, and optimization under uncertainties: a review

Replication of results  In this review paper, we do not provide any Au S (2005) Reliability-based design sensitivity by efficient simulation.
results to replicate. Comput struct 83(14):1048–1061
Au S, Beck J (2001) Estimation of small failure probabilities in
high dimensions by subset simulation. Probab Eng Mech
16(4):263–277
Au S, Papadimitriou C, Beck J (1999) Reliability of uncertain
References dynamical systems with multiple design points. Struct Saf
21(2):113–133
Acar E (2016) A reliability index extrapolation method for separable Ayyub B, McCuen R (2016) Probability, statistics, and reliability for
limit states. Struct Multidiscip Optim 53:1099–1111 engineers and scientists. CRC Press, Boca Raton
Acar E, Solanki K (2009) System reliability based vehicle design for Azarkish H, Rashki M (2019) Reliability and reliability-based sensitiv-
crashworthiness and effects of various uncertainty reduction ity analysis of shell and tube heat exchangers using Monte Carlo
measures. Struct Multidiscip Optim 39(3):311–325 simulation. Appl Therm Eng 159:113842
Acar E, Kale A, Haftka R, Stroud W (2006) Structural safety measures Azevedo CL, Ciuffo B, Cardoso JL, Ben-Akiva ME (2015) Dealing
for airplanes. J Aircr 43(1):30–38 with uncertainty in detailed calibration of traffic simulation mod-
Acar E, Haftka R, Johnson T (2007) Tradeoff of uncertainty reduc- els for safety assessment. Transp Res C 58:395–412
tion mechanisms for reducing structural weight. J Mech Des Ba-Abbad M, Nikolaidis E, Kapania R (2006) New approach for system
129(3):266–274 reliability-based design optimization. AIAA J 44(5):1087–1096
Acar E, Haftka R, Kim N (2010) Effects of structural tests on aircraft Bacarreza O, Aliabadi M, Apicella A (2015) Robust design and optimi-
safety. AIAA J 48(10):2235–2248 zation of composite stiffened panels in post-buckling.structural
Agarwal H, Mozumder C, Renaud J, Watson L (2007) An inverse- and multidisciplinary
measure-based unilevel architecture for reliability-based design Bae H, Alyanak E (2016) Sequential subspace reliability method with
optimization. Struct Multidiscip Optim 33(3):217–227 univariate revolving integration. AIAA J 54(7):2160–2170
Agarwal P, Nayal H (2015) Possibility theory versus probability theory Bashtannyk D, Hyndman R (2001) Bandwidth selection for ker-
in fuzzy measure theory. Int J Eng Res Appl 5(5):37–43 nel conditional density estimation. Comput Stat Data Anal
Ahmad I (1982) Nonparametric estimation of the location and scale 36(3):279–298
parameters based on density estimation. Ann Inst Stat Math Basudhar A, Missoum S (2008) Adaptive explicit decision functions
34(1):39–53 for probabilistic design and optimization using support vector
Akaike H (1974) A new look at the statistical model identification. machines. Comput Struct 86(19–20):1904
IEEE Trans Autom Control 19(6):716–723 Basudhar A, Missoum S, Sanchez A (2008) Limit state function
Allaire D, Noel G, Willcox K, Cointin R (2014) Uncertainty quanti- identification using support vector machines for discontinu-
fication of an aviation environmental toolsuite. Reliab Eng Syst ous responses and disjoint failure domains. Probab Eng Mech
Saf 126:14–24 23(1):1–1
Alleman G (2014) Performance-based project management: increasing Baudoui V, Klotz P, Hiriart-Urruty J, Jan S, Morel F (2012) Local
the probability of project success. Amacom uncertainty processing (LOUP) method for multidiscipli-
Allen JK, Panchal J, Mistree F, Singh AK, Gautham B (2015) Uncer- nary robust design optimization. Struct Multidiscip Optim
tainty management in the integrated realization of materials 46(5):711–726
and components. In: Proceedings of the 3rd World Congress on Bayes T (1991) An essay towards solving a problem in the doctrine of
Integrated Computational Materials Engineering (ICME 2015), chances. Comput Med Pract 8(3):157
Springer, pp 339–346 Beck J, Katafygiotis L (1998) Updating models and their uncertainties.
Allen M, Maute K (2004) Reliability-based design optimization of I: Bayesian statistical framework. J Eng Mech 124(4):455–461
aeroelastic structures. Struct Multidiscip Optim 27(4):228–242 Ben-Haim Y (1994) A non-probabilistic concept of reliability. Struct
Alyanak E, Grandhi R, Bae H (2008) Gradient projection for reliabil- Saf 14(4):227–245
ity-based design optimization using evidence theory. Eng Optim Ben-Haim Y (2001) Information-gap decision theory: decisions under
40(10):923–935 severe uncertainty. Academic Press, Cambridge
An D, Choi J (2012) Efficient reliability analysis based on Bayesian Ben-Haim Y (2006) Information-gap decision theory: decisions under
framework under input variable and metamodel uncertainties. severe uncertainty, 2nd edn. Academic Press, London
Struct Multidiscip Optim 46(4):533–547 Ben-Haim Y, Elishakoff I (1995) Discussion on: a non-probabilistic
Anderson T, Darling D (1952) Asymptotic theory of certain good- concept of reliability. Struct Saf 17(3):195–199
ness of fit criteria based on stochastic processes. Ann Math Stat Ben-Haim Y, Elishakoff I (2013) Convex models of uncertainty in
23(2):193–212 applied mechanics. Elsevier, Amsterdam
Annis C (2004) Probabilistic life prediction isn’t as easy as it looks. Benner P, Gugercin S, Willcox K (2015) A survey of projection-based
In: Johnson WS, Hillberry BM (eds) Probabilistic aspects of life model reduction methods for parametric dynamical systems.
prediction. ASTM International, West Conshohocken SIAM Rev 57(4):483–531
António CC, Hoffbauer LN (2009) An approach for reliability-based Beyer HG, Sendhoff B (2007) Robust optimization-a comprehensive
robust design optimisation of angle-ply composites. Compos survey. Comput Methods Appl Mech Eng 196(33–34):3190–3218
Struct 90(1):53–59 Bichon B, Eldred M, Swiler L, Mahadevan S, McFarland J (2008)
Aoues Y, Chateauneuf A (2010) Benchmark study of numerical meth- Efficient global reliability analysis for nonlinear implicit perfor-
ods for reliability-based design optimization. Struct Multidiscip mance functions. AIAA J 46(10):2459–2468
Optim 41(2):277–294 Bichon B, Eldred M, Mahadevan S, McFarland J (2013) Efficient
Arendt P, Apley D, Chen W (2012) Quantification of model uncer- global surrogate modeling for reliability-based design optimiza-
tainty: calibration, model discrepancy, and identifiability. J Mech tion. J Mech Des 135(1):011009
Des 134(10):100908 Blatman G (2009) Adaptive sparse polynomial chaos expansions for
Arslan A, Kaya M (2001) Determination of fuzzy logic member- uncertainty propagation and sensitivity analysis
ship functions using genetic algorithms. Fuzzy Sets Syst
118(2):297–306

13
 E. Acar et al.

Blatman G, Sudret B (2010a) An adaptive algorithm to build up sparse Chen Z, Peng S, Li X, Qiu H, Xiong H, Gao L, Li P (2015) An impor-
polynomial chaos expansions for stochastic finite element analy- tant boundary sampling method for reliability-based design
sis. Probab Eng Mech 25(2):183–197 optimization using Kriging model. Struct Multidiscip Optim
Blatman G, Sudret B (2010b) Efficient computation of global sensitiv- 52(1):55–70
ity indices using sparse polynomial chaos expansions. Reliab Eng Chen Z, Li X, Chen G, Gao L, Qiu H, Wang S (2018) A probabilistic
Syst Saf 95(11):1216–1229 feasible region approach for reliability-based design optimiza-
Booker A, Dennis J, Frank P, Serafini D, Torczon V, Trosset M (1999) tion. Struct Multidiscip Optim 57(1):359–372
A rigorous framework for optimization of expensive functions Chen Z, Wu Z, Li X, Chen G, Gao L, Gan X, Wang S (2019a) A
by surrogates. Struct Optim 17(1):1–13 multiple-design-point approach for reliability-based design opti-
Bowman A (1984) An alternative method of cross-validation for the mization. Eng Optim 51(5):875–895
smoothing of density estimates. Biometrika 71(2):353–360 Chen Z, Zhou P, Liu Y (2019b) A novel approach to uncertainty analy-
Breitung K (1984) Asymptotic approximations for multinormal inte- sis using methods of hybrid dimension reduction and improved
grals. J Eng Mech 110(3):357–366 maximum entropy. Struct Multidiscip Optim 60:1841–1866
Breitung K (2019) The geometry of limit state function graphs and Cheng H, Chen J (1997) Automatically determine the membership
subset simulation: Counterexamples. Reliab Eng Syst Saf function based on the maximum entropy principle. Inf Sci
182:98–106 96(3–4):163–182
Broemeling L (2011) An account of early statistical inference in Arab Cheng J, Liu Z, Qian Y, Zhou Z, Tan J (2020) Non-probabilistic robust
cryptology. Am Stat 65(4):255–257 equilibrium optimization of complex uncertain structures. J
Burnham K, Anderson D (2004) Multimodel inference: understand- Mech Des 142(2):021405
ing AIC and BIC in model selection. Sociol Methods Res Chiralaksanakul A, Mahadevan S (2005) First-order approximation
33(2):261–304 methods in reliability-based design optimization. J Mech Des
Cadini F, Santos F, Zio E (2014) An improved adaptive Kriging-based 127:851
importance technique for sampling multiple failure regions of Cho H, Choi K, Gaul N, Lee I, Lamb D, Gorsich D (2016a) Conserva-
low probability. Reliab Eng Syst Saf 131:109–117 tive reliability-based design optimization method with insuffi-
Cadini F, Gioletta A, Zio E (2015) Improved metamodel-based impor- cient input data. Struct Multidiscip Optim 54(6):1609–1630
tance sampling for the performance assessment of radioactive Cho H, Choi K, Lee I, Lamb D (2016b) Design sensitivity method for
waste repositories. Reliab Eng Syst Saf 134:188–197 sampling-based RBDO with varying standard deviation. J Mech
das Chagas Moura M, Zio E, Lins ID, Droguett E (2011) Failure and Des 138(1):011405
reliability prediction by support vector machines regression of Cho H, Choi K, Shin J (2020) Iterative most probable point search
time series data. Reliab Eng Syst Saf 96:1527–1534 method for problems with a mixture of random and interval vari-
Chakraborty S, Chatterjee T, Chowdhury R, Adhikari S (2017) A sur- ables. J Mech Des 142(7):071703
rogate based multi-fidelity approach for robust design optimiza- Cho S, Jang J, Kim S, Park S, Lee T, Lee M, Hong S (2016) Nonpara-
tion. Appl Math Model 47:726–744 metric approach for uncertainty-based multidisciplinary design
Chan K, Skerlos S, Papalambros P (2007) An adaptive sequential linear optimization considering limited data. Struct Multidiscip Optim
programming algorithm for optimal design problems with proba- 54(6):1671–1688
bilistic constraints. J Mech Des 129(2):140–149 Cho T, Lee B (2011) Reliability-based design optimization using
Chatterjee T, Chakraborty S, Chowdhury R (2019) A critical review convex linearization and sequential optimization and reliability
of surrogate assisted robust design optimization. Arch Comput assessment method. Struct Saf 33(1):42–50
Methods Eng 26(1):245–274 Chutia R (2017) Uncertainty quantification under hybrid structure of
Chaudhuri A, Haftka R (2013) Separable Monte Carlo combined with probability-fuzzy parameters in Gaussian plume model. Life
importance sampling for variance reduction. Int J Reliab Saf Cycle Reliab Saf Eng 6(4):277–284
7(3):201–215 Cicala D, Irias X (2014) Utilizing info-gap decision theory to improve
Chaudhuri A, Kramer B, Willcox K (2020) Information reuse for pipeline reliability: a case study. In: Pipelines 2014: from under-
importance sampling in reliability-based design optimization. ground to the forefront of innovation and sustainability, pp
Reliab Eng Syst Saf 201:106853 1749–1760
Chen G, Fan J, Xu H, Li B (2020) Calculation of hybrid reliability of Civanlar M, Trussell H (1986) Constructing membership functions
turbine disk based on self-evolutionary game model with few using statistical data. Fuzzy Sets Syst 18(1):1–13
shot learning. Struct Multidiscip Optim 2020:1–13 Constantine P, Emory M, Larsson J, Iaccarino G (2015) Exploiting
Chen S, Yang X (2000) Interval finite element method for beam struc- active subspaces to quantify uncertainty in the numerical simula-
tures. Finite Elem Anal Des 34(1):75–88 tion of the HyShot II scramjet. J Comput Phys 302:1–20
Chen S, Nikolaidis E, Cudney H, Rosca R, Haftka R (1999) Compari- Coppitters D, De Paepe W, Contino F (2019) Surrogate-assisted robust
son of probabilistic and fuzzy set methods for designing under design optimization and global sensitivity analysis of a directly
uncertainty. In: 40th structures, structural dynamics, and materi- coupled photovoltaic-electrolyzer system under techno-economic
als conference and exhibit, p 1579 uncertainty. Appl Energy 248:310–320
Chen X, Hasselman T, Neill D (1997) Reliability-based structural Council NR et al (2009) Science and decisions: advancing risk assess-
design optimization for practical applications. In: Proceedings ment. National Academies Press, Washington DC
of the 38th AIAA structures, structural dynamics, and materials Degrauwe D, Lombaert G, De Roeck G (2010) Improving interval
conference, Florida analysis in finite element calculations by means of affine arith-
Chen Z, Qiu H, Gao L, Su L, Li P (2013) An adaptive decoupling metic. Comput Struct 88(3–4):247–254
approach for reliability-based design optimization. Comput Dempster A (1967) Upper and lower probabilities induced by a multi-
Struct 117:58–66 valued mapping. Ann Math Stat 38(2):325–339
Chen Z, Qiu H, Gao L, Li X, Li P (2014) A local adaptive sampling Der Kiureghian A (1996) Structural reliability methods for seismic
method for reliability-based design optimization using Kriging safety assessment: a review. Eng Struct 18(6):412–424
model. Struct Multidiscip Optim 49(3):401–416 Der Kiureghian A, Dakessian T (1998) Multiple design points in first
and second-order reliability. Struct Saf 20(1):37–49

13
Modeling, analysis, and optimization under uncertainties: a review

Dodson M, Parks G (2015) Robust aerodynamic design optimization Fan X, Wang P, Hao F (2019) Reliability-based design optimization
using polynomial chaos. J Aircr 46(2):635–646 of crane bridges using Kriging-based surrogate models. Struct
Doltsinis I, Kang Z (2004) Robust design of structures using Multidiscip Optim 59(3):993–1005
optimization methods. Comput Methods Appl Mech Eng Fang J, Gao Y, Sun G, Xu C, Li Q (2015) Multiobjective robust design
193(23–26):2221–2237 optimization of fatigue life for a truck cab. Reliab Eng Syst Saf
Du L, Choi K, Youn B, Gorsich D (2006) Possibility-based design 135:1–8
optimization method for design problems with both statistical Ferson S, Ginzburg L (1996) Different methods are needed to propagate
and fuzzy input data. J Mech Des 128(4):928 ignorance and variability. Reliab Eng Syst Saf 54(2–3):133–144
Du X, Chen W (2004) Sequential optimization and reliability assess- Ferson S, Joslyn C, Helton J, Oberkampf W, Sentz K (2004) Summary
ment method for efficient probabilistic design. J Mech Des from the epistemic uncertainty workshop: consensus amid diver-
126(2):225 sity. Reliab Eng Syst Saf 85(1–3):355–369
Du X, Hu Z (2012) First order reliability method with truncated ran- Freund R (2003) Model reduction methods based on Krylov subspaces.
dom variables. J Mech Des 134(9):091005 Acta Numer 12:267–319
Du X, Sudjianto A, Chen W (2004) An integrated framework for opti- Gao W, Wu D, Song C, Tin-Loi F, Li X (2011) Hybrid probabilis-
mization under uncertainty using inverse reliability strategy. J tic interval analysis of bar structures with uncertainty using a
Mech Des 126(4):562–570 mixed perturbation Monte-Carlo method. Finite Elem Anal Des
Du X, Sudjianto A, Huang B (2005) Reliability-based design with 47(7):643–652
the mixture of random and interval variables. J Mech Des Gersem D, Hilde DM, Desmet W, Vandepitte D (2006) Non-prob-
127(6):1068 abilistic uncertainty assessment in finite element models with
Duan Z, Jung Y, Yan J, Lee I (2020) Reliability-based multi-scale superelements. In: 47th AIAA/ASME/ASCE/AHS/ASC struc-
design optimization of composite frames considering structural tures, structural dynamics, and materials conference 14th AIAA/
compliance and manufacturing constraints. Struct Multidiscip ASME/AHS adaptive structures conference 7th, p 2072
Optim 61(6):2401–2421 Ghanem R, Spanos P (1991) Stochastic finite element method: response
Dubois D, Prade H (1988) Possibility theory. Plenum, New York statistics. Stochastic finite elements: a spectral approach.
Dubourg V, Sudret B, Bourinet J (2011) Reliability-based design opti- Springer, New York, pp 101–119
mization using Kriging surrogates and subset simulation. Struct Ghanem R, Higdon D, Owhadi H (2017) Handbook of uncertainty
Multidiscip Optim 44(5):673–690 quantification. Springer, New York
Dubourg V, Sudret B, Deheeger F (2013) Metamodel-based importance Ghisu T, Parks GT, Jarrett JP, Clarkson PJ (2011) Robust design opti-
sampling for structural reliability analysis. Probab Eng Mech mization of gas turbine compression systems. J Propul Power
33:47–57 27(2):282–295
Duong P, Yang Q, Park H, Raghavan N (2019) Reliability analysis and Giles M (2008) Multilevel Monte Carlo path simulation. Oper Res
design of a single diode solar cell model using polynomial chaos 56(3):607–617
and active subspace. Microelectron Reliab 100:113477 Goel T, Haftka R, Shyy W, Queipo N (2007) Ensemble of surrogates.
Duong T, Hazelton M (2003) Plug-in bandwidth matrices for bivariate Struct Multidiscip Optim 33(3):199–216
kernel density estimation. J Nonparametr Stat 15(1):17–30 Gomes HM, Awruch AM, Lopes PAM (2011) Reliability based opti-
Duong T, Hazelton M (2005) Cross-validation bandwidth matri- mization of laminated composite structures using genetic algo-
ces for multivariate kernel density estimation. Scand J Stat rithms and artificial neural networks. Struct Saf 33(3):186–195
32(3):485–506 Grujicic M, Arakere G, Bell W, Marvi H, Yalavarthy H, Pandurangan
Echard B, Gayton N, Lemaire M, Relun N (2013) A combined impor- B, Haque I, Fadel G (2010) Reliability-based design optimization
tance sampling and Kriging reliability method for small failure for durability of ground vehicle suspension system components.
probabilities with time-demanding numerical models. Reliab Eng J Mater Eng Perform 19(3):301–313
Syst Saf 111:232–240 Guo S, Lu Z (2015) A non-probabilistic robust reliability method for
El Moçayd N, Mohamed M, Ouazar D, Seaid M (2020) Stochastic analysis and design optimization of structures with uncertain-
model reduction for polynomial chaos expansion of acoustic but-bounded parameters. Appl Math Model 39(7):1985–2002
waves using proper orthogonal decomposition. Reliab Eng Syst Guyonnet D, Bourgine B, Dubois D, Fargier H, Co me B, Chilès JP,
Saf 195:106733 (2003) Hybrid approach for addressing uncertainty in risk assess-
Elishakoff I, Bekel Y (2013) Application of Lame’s super ellipsoids to ments. J Environ Eng 129(1):68–78
model initial imperfections. J Appl Mech 80(6): 061006 Hájek A (2019) Interpretations of probability, the stanford encyclope-
Elishakoff I, Zingales M (2003) Contrasting probabilistic and anti- dia of philosophy
optimization approaches in an applied mechanics problem. Int J Håkansson A (2019) Estimating convective heat transfer coefficients
Solids Struct 40(16):4281–4297 and uncertainty thereof using the general uncertainty manage-
Elishakoff I, Elisseeff P, Glegg S (1994a) Nonprobabilistic, convex- ment (GUM) framework. J Food Eng 263:53–62
theoretic modeling of scatter in material properties. AIAA J Hao P, Wang Y, Liu C, Wang B, Wu H (2017) A novel non-prob-
32(4):843–849 abilistic reliability-based design optimization algorithm using
Elishakoff I, Haftka R, Fang J (1994b) Structural design under bounded enhanced chaos control method. Comput Methods Appl Mech
uncertainty-optimization with anti-optimization. Comput Struct Eng 318:572–593
53(6):1401–1405 Hasofer A, Lind N (1974) Exact and invariant second-moment code
Ellingwood B (1980) Development of a probability based load criterion format. J Eng Mech Div 100(1):111–121
for American National Standard A58: Building code require- Hassan R, Crossley W (2008) Spacecraft reliability-based design
ments for minimum design loads in buildings and other struc- optimization under uncertainty including discrete variables. J
tures, vol 13. National Bureau of Standards, US Department of Spacecr Rocket 45(2):394–405
Commerce Hasuike T, Katagiri H (2016) Construction of an appropriate member-
Engelund S, Rackwitz R (1993) A benchmark study on impor- ship function based on size of fuzzy set and mathematical pro-
tance sampling techniques in structural reliability. Struct Saf gramming. In: Proceedings of the international multiconference
12:255–276 of engineers and computer scientists, vol 2

13
 E. Acar et al.

Hawchar L, El Soueidy CP, Schoefs F (2018) Global Kriging surro- Isight (2021) Simulia execution engine—-dassault systèmesⓇ . https://​
gate modeling for general time-variant reliability-based design www.​3ds.​com/​produ​cts-​servi​ces/​simul​ia/​produ​cts/​isight-​simul​
optimization problems. Struct Multidiscip Optim 58:955–968 ia-​execu​tion-​engine/
He W, Zeng Y, Li G (2020) An adaptive polynomial chaos expan- Ito M, Kim N, Kogiso N (2018) Conservative reliability index for epis-
sion for high-dimensional reliability analysis. Struct Multidiscip temic uncertainty in reliability-based design optimization. Struct
Optim 62:2051–2067 Multidiscip Optim 57(5):1919–1935
Helton J, Davis F (2003) Latin hypercube sampling and the propaga- Jalota H, Thakur M, Mittal G (2017) A credibilistic decision support
tion of uncertainty in analyses of complex systems. Reliab Eng system for portfolio optimization. Appl Soft Comput 59:512–528
Syst Saf 81(1):23–69 Jang J (1993) Anfis: adaptive-network-based fuzzy inference system.
Helton JC, Johnson JD, Sallaberry CJ, Storlie CB (2006) Survey of IEEE Trans Syst Man Cybern 23(3):665–685
sampling-based methods for uncertainty and sensitivity analysis. Jensen H, Valdebenito M, Schuëller G, Kusanovic D (2009) Reliability-
Reliab Eng Syst Saf 91(10–11):1175–1209 based optimization of stochastic systems using line search. Com-
Hess P, Bruchman D, Assakkaf I, Ayyub B (2002) Uncertainties in put Methods Appl Mech Eng 198(49–52):3915–3924
material and geometric strength and load variables. Nav Eng J Jeong S, Park G (2017) Single loop single vector approach using the
114(2):139–166 conjugate gradient in reliability based design optimization. Struct
Hoffman F, Hammonds J (1994) Propagation of uncertainty in risk Multidiscip Optim 55(4):1329–1344
assessments: the need to distinguish between uncertainty due to Ji W, Ren Z, Marzouk Y, Law C (2019) Quantifying kinetic uncertainty
lack of knowledge and uncertainty due to variability. Risk Anal in turbulent combustion simulations using active subspaces. Proc
14(5):707–712 Combust Inst 37(2):2175–2182
Hong T, Lee C (1996) Induction of fuzzy rules and membership func- Jiang C, Han X, Li W, Liu J, Zhang Z (2012a) A hybrid reliability
tions from training examples. Fuzzy Sets Syst 84(1):33–47 approach based on probability and interval for uncertain struc-
Hora S (1996) Aleatory and epistemic uncertainty in probability tures. J Mech Des 134(3):031001
elicitation with an example from hazardous waste management. Jiang C, Lu G, Han X, Liu L (2012b) A new reliability analysis
Reliab Eng Syst Saf 54(2–3):217–223 method for uncertain structures with random and interval vari-
Hosder S, Watson L, Grossman B, Mason W, Kim H, Haftka R, ables. Int J Mech Mater Des 8(2):012–9184
Cox S (2001) Polynomial response surface approximations for Jiang C, Bi R, Lu G, Han X (2013a) Structural reliability analysis
the multidisciplinary design optimization of a high speed civil using non-probabilistic convex model. Comput Methods Appl
transport. Optim Eng 2(4):431–452 Mech Eng 254:83–98
Hoseyni S, Pourgol-Mohammad M, Tehranifard A, Yousefpour Jiang C, Zhang Q, Han X, Liu J, Hu D (2015) Multidimensional
F (2014) A systematic framework for effective uncertainty parallelepiped model-a new type of non-probabilistic convex
assessment of severe accident calculations; hybrid qualitative model for structural uncertainty analysis. Int J Numer Methods
and quantitative methodology. Reliab Eng Syst Saf 125:22–35 Eng 103(1):31–59
Hosseinzadeh Y, Taghizadieh N, Jalili S (2018) A new structural rea- Jiang C, Qiu H, Gao L, Cai X, Li P (2017) An adaptive hybrid
nalysis approach based on the polynomial-type extrapolation single-loop method for reliability-based design optimization
methods. Struct Multidiscip Optim 58(3):1033–1049 using iterative control strategy. Struct Multidiscip Optim
Hu C, Youn BD (2011) Adaptive-sparse polynomial chaos expan- 56(6):1271–1286
sion for reliability analysis and design of complex engineering Jiang C, Hu Z, Liu Y, Mourelatos ZP, Gorsich D, Jayakumar P (2020)
systems. Struct Multidiscip Optim 43(3):419–442 A sequential calibration and validation framework for model
Hu W, Choi K, Cho H (2016) Reliability-based design optimization uncertainty quantification and reduction. Comput Methods Appl
of wind turbine blades for fatigue life under dynamic wind load Mech Eng 368:113172
uncertainty. Struct Multidiscip Optim 54(4):953–970 Jiang H, Deng H, He Y (2008) Determination of fuzzy logic member-
Hu X, Parks G, Chen X, Seshadri P (2015) Discovering a one- ship functions using extended ant colony optimization algorithm.
dimensional active subspace to quantify multidisciplinary In: 2008 Fifth international conference on fuzzy systems and
uncertainty in satellite system design. Adv Space Res 57:1268 knowledge discovery, IEEE, vol 1, pp 581–585
Hu X, Chen X, Zhao Y, Tuo Z, Yao W (2017) Active subspace Jiang Z, Li J (2017) High dimensional structural reliability with dimen-
approach to reliability and safety assessments of small satel- sion reduction. Struct Saf 69:35–46
lite separation. Acta Astronaut 131:159–165 Jiang Z, Chen W, Fu Y, Yang R (2013b) Reliability-based design opti-
Hu Z, Du X (2013a) A sampling approach to extreme value dis- mization with model bias and data uncertainty. SAE Int J Mater
tribution for time-dependent reliability analysis. J Mech Des Manuf 6(3):502–516
135:071003 Jiao G, Moan T (1990) Methods of reliability model updating through
Hu Z, Du X (2013b) Time-dependent reliability analysis with joint additional events. Struct Saf 9(2):139–153
up-crossing rates. Struct Multidiscip Optim 48:893–907 Jo H, Lee K, Lee M, Jung Y, Lee I (2021) Optimization-based model
Hu Z, Du X (2015) First order reliability method for time-variant calibration of marginal and joint output distributions utilizing
problems using series expansions. Struct Multidiscip Optim analytical gradients. Struct Multidiscip Optim 63:1–16
51:1–21 Ju B, Lee B (2008) Reliability-based design optimization using
Huang B, Du X (2006) Uncertainty analysis by dimension reduction a moment method and a Kriging metamodel. Eng Optim
integration and saddlepoint approximations 40(5):421–438
Huang X, Li Y, Zhang Y, Zhang X (2018) A new direct second-order Jung Y, Cho H, Lee I (2019a) MPP-based approximated DRM (ADRM)
reliability analysis method. Appl Math Model 55:68–80 using simplified bivariate approximation with linear regression.
Huang Z, Jiang C, Zhou Y, Luo Z, Zhang Z (2016) An incremental Struct Multidiscip Optim 59(5):1761–1773
shifting vector approach for reliability-based design optimiza- Jung Y, Cho H, Lee I (2019b) Reliability measure approach for con-
tion. Struct Multidiscip Optim 53(3):523–543 fidence-based design optimization under insufficient input data.
Iooss B, Le Gratiet L (2019) Uncertainty and sensitivity analysis of Struct Multidiscip Optim 60(5):1967–1982
functional risk curves based on Gaussian processes. Reliab Eng Jung Y, Cho H, Duan Z, Lee I (2020a) Determination of sample size for
Syst Saf 187:58–66 input variables in RBDO through bi-objective confidence-based

13
Modeling, analysis, and optimization under uncertainties: a review

design optimization under input model uncertainty. Struct Multi- Kolmogorov A (1933) Sulla determinazione empirica di une legge
discip Optim 61(1):253–266 di distribuzione. Giornale dell’Istituto Italiano degli Attuari
Jung Y, Cho H, Lee I (2020b) Intelligent initial point selection for MPP 4:83–91
search in reliability-based design optimization. Struct Multidis- Konečná K, Horová I (2019) Maximum likelihood method for band-
cip Optim 62:1–12 width selection in kernel conditional density estimate. Comput
Jung Y, Kang K, Cho H, Lee I (2021) Confidence-based design opti- Stat 34(4):1871–1887
mization for a more conservative optimum under surrogate Kumar R, Ali S, Jeyaraman S, Gupta S (2020) Uncertainty quantifica-
model uncertainty caused by gaussian process. J Mech Des tion of bladed disc systems using data driven stochastic reduced
143(9):091701 order models. Int J Mech Sci 190:106011
Kale A, Haftka R (2008) Tradeoff of weight and inspection cost in Kumar S, Pippy R, Acar E, Kim N, Haftka R (2009) Approximate
reliability-based structural optimization. J Aircr 45(1):77–85 probabilistic optimization using exact-capacity-approximate-
Kang HY, Kwak BM (2009) Application of maximum entropy princi- response-distribution (ECARD). Struct Multidiscip Optim
ple for reliability-based design optimization. Struct Multidiscip 38:613–626
Optim 38(4):331–346 Laplace P (1812) Analytical theory of probability. Courier, Paris
Kang K, Qin C, Lee B, Lee I (2019) Modified screening-based Krig- Lee D, Kim N, Kim H (2016) Validation and updating in a large auto-
ing method with cross validation and application to engineering motive vibro-acoustic model using a P-box in the frequency
design. Appl Math Model 70:626–642 domain. Springer-Verlag, New York
Kang S, Park J, Lee I (2017a) Accuracy improvement of the most prob- Lee G, Kim W, Oh H, Youn B, Kim N (2019a) Review of statistical
able point-based dimension reduction method using the Hessian model calibration and validation-from the perspective of uncer-
matrix. Int J Numer Methods Eng 111(3):203–217 tainty structures. Struct Multidiscip Optim 60(4):1619–1644
Kang Y, Hong J, Lim O, Noh Y (2017b) Reliability analysis using Lee I, Choi K, Du L, Gorsich D (2008a) Dimension reduction method
parametric and nonparametric input modeling methods. J Com- for reliability-based robust design optimization. Comput Struct
put Struct Eng Inst Korea 30(1):87–94 86(13–14):1550–1562
Kang Y, Noh Y, Lim O (2018) Kernel density estimation with bounded Lee I, Choi K, Du L, Gorsich D (2008b) Inverse analysis method using
data. Struct Multidiscip Optim 57(1):95–113 MPP-based dimension reduction for reliability-based design opti-
Kang Y, Noh Y, Lim O (2019) Integrated statistical modeling method: mization of nonlinear and multi-dimensional systems. Comput
part I-statistical simulations for symmetric distributions. Struct Methods Appl Mech Eng 198(1):14–27
Multidiscip Optim 60(5):1719–1740 Lee I, Choi K, Gorsich D (2010) System reliability-based design opti-
Kang Z, Bai S (2013) On robust design optimization of truss struc- mization using the MPP-based dimension reduction method.
tures with bounded uncertainties. Struct Multidiscip Optim Struct Multidiscip Optim 41(6):823–839
47(5):699–714 Lee I, Choi K, Noh Y, Zhao L, Gorsich D (2011) Sampling-based
Kang Z, Luo Y, Li A (2011) On non-probabilistic reliability-based stochastic sensitivity analysis using score functions for RBDO
design optimization of structures with uncertain-but-bounded problems with correlated random variables. J Mech Des. https://​
parameters. Struct Saf 33(3):196–205 doi.​org/​10.​1115/​DETC2​010-​28591
Kanno Y (2019) A data-driven approach to non-parametric reliability- Lee I, Noh Y, Yoo D (2012) A novel second-order reliability method
based design optimization of structures with uncertain load. (SORM) using noncentral or generalized chi-squared distribu-
Struct Multidiscip Optim 60(1):83–97 tions. J Mech Des 134(10):100912
Kanno Y, Takewaki I (2006) Robustness analysis of trusses with Lee I, Choi K, Noh Y, Lamb D (2013) Comparison study between
separable load and structural uncertainties. Int J Solids Struct probabilistic and possibilistic methods for problems under a lack
43(9):2646–2669 of correlated input statistical information. Struct Multidiscip
Kaufman J, Prager M (1990) Marine structural steel toughness data Optim 47(2):175–189
bank. In: National materials property data network, Columbus Lee J, Kwak B (1995) Reliability-based structural optimal design
OH, abridged edn using the Neumann expansion technique. Comput Struct
Kaymaz I, McMahon C (2005) A response surface method based on 55(2):287–296
weighted regression for structural reliability analysis. Probab Eng Lee KH, Park GJ (2001) Robust optimization considering tolerances
Mech 20:11–17 of design variables. Comput Struct 79(1):77–86
Keane AJ, Voutchkov II (2020) Robust design optimization using sur- Lee S, Chen W (2009) A comparative study of uncertainty propaga-
rogate models. J Comput Des Eng 7(1):44–55 tion methods for black-box-type problems. Struct Multidiscip
Kennedy MC, O’Hagan A (2001) Bayesian calibration of computer Optim 37(3):239
models. J R Stat Soc 63(3):425–464 Lee T, Jung J (2008) A sampling technique enhancing accuracy and
Keshtegar B, Hao P (2017) A hybrid self-adjusted mean value method efficiency of metamodel-based RBDO: constraint boundary sam-
for reliability-based design optimization using sufficient descent pling. Comput Struct 86(13–14):1463–1476
condition. Appl Math Model 41:257–270 Lee U, Kang N, Lee I (2019) Selection of optimal target reliability
Kim N, Wang H, Queipo N (2006) Efficient shape optimization under in RBDO through reliability-based design for market systems
uncertainty using polynomial chaos expansions and local sensi- (RBDMS) and application to electric vehicle design. Struct
tivities. AIAA J 44(5):1112–1116 Multidiscip Optim 60(3):949–963
Kim T, Lee G, Youn B (2019) Uncertainty characterization under Lee U, Kang N, Lee I (2020a) Shared autonomous electric vehicle
measurement errors using maximum likelihood estimation: design and operations under uncertainties: a reliability-based
cantilever beam end-to-end UQ test problem. Struct Multidiscip design optimization approach. Struct Multidiscip Optim
Optim 59(2):323–333 61(4):1529–1545
Knight FH (1921) Risk, uncertainty and profit, vol 31. Houghton Lee U, Park S, Lee I (2020b) Robust design optimization (rdo) of ther-
Mifflin, Boston moelectric generator system using non-dominated sorting genetic
Kolmogoroff A (1941) Confidence limits for an unknown distribution algorithm II (nsga-II). Energy 196:117090
function. Ann Math Stat 12(4):461–463 Li G, Zhang K (2011) A combined reliability analysis approach with
dimension reduction method and maximum entropy method.
Struct Multidiscip Optim 43:121–134

13
 E. Acar et al.

Li H, Cao Z (2016) Matlab codes of subset simulation for reliability the structures with interval uncertainty. Int J Mech Mater Des
analysis and structural optimization. Struct Multidiscip Optim 12(3):375–386
54(2):391–410 Liu P, Der Kiureghian A (1991) Optimization algorithms for structural
Li H, Cho H, Sugiyama H, Choi K, Gaul NJ (2017) Reliability-based reliability. Struct Saf 9:161–178
design optimization of wind turbine drivetrain with integrated Liu X, Wu Y, Wang B, Ding J, Jie H (2017) An adaptive local range
multibody gear dynamics simulation considering wind load sampling method for reliability-based design optimization using
uncertainty. Struct Multidiscip Optim 56(1):183–201 support vector machine and Kriging model. Struct Multidiscip
Li J, Jiang C, Ni B, Zhan L (2019a) Uncertain vibration analysis based Optim 55(6):2285–2304
on the conceptions of differential and integral of interval process. Lopez RH, Lemosse D, de Cursi JES, Rojas J, El-Hami A (2011) An
Int J Mech Mater Des 16:225 approach for the reliability based design optimization of lami-
Li L, Wan H, Gao W, Tong F, Li H (2019b) Reliability based multi- nated composites. Eng Optim 43(10):1079–1094
disciplinary design optimization of cooling turbine blade con- Luo Z, Wang X, Shi Q, Liu D (2021) Ubc-constrained non-probabilistic
sidering uncertainty data statistics. Struct Multidiscip Optim reliability-based optimization of structures with uncertain-but-
59(2):659–673 bounded parameters. Struct Multidiscip Optim 63(1):311–326
Li M, Wang Z (2018) Confidence-driven design optimization using Madsen H, Krenk S, Lind N (2006) Methods of structural safety. Cou-
Gaussian process metamodeling with insufficient data. J Mech rier Corporation
Des 140(12):121405 Mahadevan S, Zhang R, Smith N (2001) Bayesian networks for system
Li M, Wang Z (2019) Surrogate model uncertainty quantification reliability reassessment. Struct Saf 23(3):231–251
for reliability-based design optimization. Reliab Eng Syst Saf Makhloufi A, Aoues Y, El Hami A (2016) Reliability based design
192:106432 optimization of wire bonding in power microelectronic devices.
Li M, Wang Z (2020) Deep learning for high-dimensional reliability Microsyst Technol 22(12):2737–2748
analysis. Mech Syst Signal Process 139:106399 Mansour R, Olsson M (2014) A closed-form second-order reliability
Li W, Gao L, Xiao M (2020) Multidisciplinary robust design opti- method using noncentral chi-squared distributions. J Mech Des
mization under parameter and model uncertainties. Eng Optim 136(10):10402
52(3):426–445 Marelli S, Sudret B (2014) Uqlab: A framework for uncertainty quan-
Li X, Qiu H, Chen Z, Gao L, Shao X (2016) A local Kriging approxi- tification in matlab. The 2nd International conference on vul-
mation method using MPP for reliability-based design optimi- nerability and risk analysis and management (ICVRAM 2014).
zation. Comput Struct 162:102–115 University of Liverpool, United Kingdom, pp 2554–2563
Li X, Gong C, Gu L, Jing Z, Fang H, Gao R (2019c) A reliabil- Martin N, England J (1981) Mathematical theory of entropy. Addi-
ity-based optimization method using sequential surrogate son-Wesley, Reading
model and Monte Carlo simulation. Struct Multidiscip Optim McAllister CD, Simpson TW (2003) Multidisciplinary robust design
59(2):439–460 optimization of an internal combustion engine. J Mech Des
Li X, Meng Z, Chen G, Yang D (2019d) A hybrid self-adjusted single- 125(1):124–130
loop approach for reliability-based design optimization. Struct McDonald M, Mahadevan S (2008) Design optimization with sys-
Multidiscip Optim 60(5):1867–1885 tem-level reliability constraints. J Mech Des 130(2):021403
Li Y, Chen J, Feng L (2012) Dealing with uncertainty: a sur- McFarland J, Mahadevan S (2008) Error and variability characteriza-
vey of theories and practices. IEEE Trans Knowl Data Eng tion in structural dynamics modeling. Comput Methods Appl
25(11):2463–2482 Mech Eng 197(29–32):2621–2631
Liang J, Mourelatos Z, Nikolaidis E (2007) A single-loop approach Melchers R (1989) Importance sampling in structural systems. Struct
for system reliability-based design optimization. J Mech Desi Saf 6:3–10
129(12):1215 Meng D, Li Y, Huang H, Wang Z, Liu Y (2015a) Reliability-based
Liang J, Mourelatos Z, Tu J (2008) A single-loop method for relia- multidisciplinary design optimization using subset simula-
bility-based design optimisation. Int J Prod Dev 5(1–2):76–92 tion analysis and its application in the hydraulic transmission
Lim J, Lee B, Lee I (2014) Second-order reliability method-based mechanism design. J Mech Des 137(5):051402
inverse reliability analysis using Hessian update for accurate and Meng Z, Li G, Wang B, Hao P (2015b) A hybrid chaos control
efficient reliability-based design optimization. Int J Numer Meth approach of the performance measure functions for reliability-
Eng 100(10):773–792 based design optimization. Comput Struct 146:32–43
Lin P, Gea HC, Jaluria Y (2011) A modified reliability index Meng Z, Zhou H, Li G, Yang D (2016) A decoupled approach for
approach for reliability-based design optimization. J Mech Des non-probabilistic reliability-based design optimization. Com-
133(4):044501 put Struct 175:65–73
Lin Q, Xiong F, Wang F, Yang X (2020) A data-driven polynomial Meng Z, Zhang D, Liu Z, Li G (2018) An adaptive directional bound-
chaos method considering correlated random variables. Struct ary sampling method for efficient reliability-based design opti-
Multidiscip Optim 62(4):2131–2147 mization. J Mech Des 140(12):121406
Liu H, Jiang C, Jia X, Long X, Zhang Z, Guan F (2018a) A new uncer- Meng Z, Zhang D, Li G, Yu B (2019) An importance learning
tainty propagation method for problems with parameterized method for non-probabilistic reliability analysis and optimi-
probability-boxes. Reliab Eng Syst Saf 172:64–73 zation. Struct Multidiscip Optim 59(4):1255–1271
Liu H, Ong Y, Cai J (2018b) A survey of adaptive sampling for global Mischke CR (1987) Prediction of stochastic endurance strength. J
metamodeling in support of simulation-based complex engineer- Vib Acoust Stress Reliab Des 109(1):113–114
ing design. Struct Multidiscip Optim 57(1):393–416 modeFRONTIER (2021) Robust design and reliability—- www.
Liu H, Jiang C, Liu J (2019) Uncertainty propagation analysis using esteco.com. https:// ​ w ww. ​ e steco. ​ c om/ ​ t echn ​ o logy/ ​ r obust-​
sparse grid technique and saddlepoint approximation based on design-​and-​relia​bility/
parameterized p-box representation. Struct Multidiscip Optim Moens D, Vandepitte D (2005) A survey of non-probabilistic uncer-
59:61–74 tainty treatment in finite element analysis. Comput Methods
Liu J, Sun X, Meng X, Li K, Zeng G, Wang X (2016) A novel Appl Mech Eng 194(12–16):1527–1555
shape function approach of dynamic load identification for

13
Modeling, analysis, and optimization under uncertainties: a review

Mohsine A, Kharmanda G, El-Hami A (2006) Improved hybrid Noh Y, Choi K, Lee I, Gorsich D, Lamb D (2011b) Reliability-based
method as a robust tool for reliability-based design optimiza- design optimization with confidence level under input model
tion. Struct Multidiscip Optim 32(3):203–213 uncertainty due to limited test data. Struct Multidiscip Optim
Moon M, Choi K, Cho H, Gaul N, Lamb D, Gorsich D (2017) 43(4):443–458
Reliability-based design optimization using confidence-based Oberguggenberger M, Fellin W (2008) Reliability bounds through ran-
model validation for insufficient experimental data. J Mech dom sets: nonparametric methods and geotechnical applications.
Des 139(3):031404 Comput Struct 86(10):1093–110
Moon M, Cho H, Choi K, Gaul N, Lamb D, Gorsich D (2018) Oberkampf W, DeLand S, Rutherford B, Diegert K, Alvin K (2002)
Confidence-based reliability assessment considering limited Error and uncertainty in modeling and simulation. Reliab Eng
numbers of both input and output test data. Struct Multidiscip Syst Saf 75(3):333–357
Optim 57(5):2027–2043 Olivier GDABCMVLA, Shields M (2020) Uqpy: a general purpose
Moon M, Choi K, Gaul N, Lamb D (2019) Treating epistemic python package and development environment for uncertainty
uncertainty using bootstrapping selection of input distribution quantification. J Comput Sci 47:101204
model for confidence-based reliability assessment. J Mech Des. Omizegba E, Adebayo G (2009) Optimizing fuzzy membership func-
https://​doi.​org/​10.​1115/1.​40421​49 tions using particle swarm algorithm. In: 2009 IEEE interna-
Moore R (1966) Interval analysis, vol 4. Prentice-Hall, Englewood tional conference on systems. man and cybernetics, IEEE, pp
Cliffs 3866–3870
Moore RE, Kearfott RB, Cloud MJ (2009) Introduction to interval OmniQuest (2021) Fesoftware. https://​omniq​uest.​world/
analysis. SIAM OptiSLang (2021) Ansys optislang. https://​www.​ansys.​com/​en-​in/​
Motta RdS, Afonso SM (2016) An efficient procedure for structural produ​cts/​platf​orm/​ansys-​optis​lang/
reliability-based robust design optimization. Struct Multidiscip Paiva R, Crawford C, Suleman A (2014) Robust and reliability-
Optim 54(3):511–530 based design optimization framework for wing design. AIAA
Mourelatos Z, Zhou J (2006) A design optimization method using J 52(4):711–724
evidence theory. J Mech Des 128(4):901 Pan H, Xi Z, Yang R (2016) Model uncertainty approximation using a
Muhanna R, Mullen R, Zhang H (2005) Penalty-based solu- copula-based approach for reliability based design optimization.
tion for the interval finite-element methods. J Eng Mech Struct Multidiscip Optim 54(6):1543–1556
131(10):1102–1111 Papaioannou I, Betz W, Zwirglmaier K, Straub D (2015) MCMC algo-
Mukhopadhyay S, Khodaparast H, Adhikari S (2016) Fuzzy uncer- rithms for subset simulation. Probab Eng Mech 41:89–103
tainty propagation in composites using gram-schmidt polynomial Papaioannou I, Breitung K, Straub D (2018) Reliability sensitivity
chaos expansion. Appl Math Model 40(7–8):4412–4428 estimation with sequential importance sampling. Struct Saf
Nagel J, Rieckermann J, Sudret B (2020) Principal component analysis 75:24–34
and sparse polynomial chaos expansions for global sensitivity Park J, Lee I (2018) A study on computational efficiency improve-
analysis and model calibration: application to urban drainage ment of novel SORM using the convolution integration. J Mech
simulation. Reliab Eng Syst Saf 195:106737 Des 140(2):025401
Nannapaneni S, Hu Z, Mahadevan S (2016) Uncertainty quantifica- Park J, Cho H, Lee I (2020) Selective dimension reduction method
tion in reliability estimation with limit state surrogates. Struct (DRM) to enhance accuracy and efficiency of most prob-
Multidiscip Optim 54(6):1509–1526 able point (MPP)-based DRM. Struct Multidiscip Optim
Nataf A (1962) Determination des distribution don’t les marges 61(3):999–1010
sont donnees. Comptes Rendus de l Academie des Sciences Parsons S, Hunter A (1998) A review of uncertainty handling for-
225:42–43 malisms. In: Applications of uncertainty formalisms, Springer,
das Neves Carneiro G, António CC, (2019) Reliability-based robust pp 8–37
design optimization with the reliability index approach applied Paté-Cornell M (1996) Uncertainties in risk analysis: six levels of
to composite laminate structures. Compos Struct 209:844–855 treatment. Reliab Eng Syst Saf 54(2–3):95–111
Ng L, Willcox K (2014) Multifidelity approaches for optimization Paulson J, Buehler E, Mesbah A (2017) Arbitrary polynomial chaos
under uncertainty. Int J Numer Meth Eng 100(10):746–772 for uncertainty propagation of correlated random variables in
Nguyen T, Song J, Paulino G (2010) Single-loop system reliability- dynamic systems. IFAC-PapersOnLine 50(1):3548–3553
based design optimization using matrix-based system reliability Pearl J (2014) Probabilistic reasoning in intelligent systems: net-
method: theory and applications. J Mech Des 132(1):011005 works of plausible inference. Elsevier, Amsterdam
Nikbay M, Kuru M (2013) Reliability based multidisciplinary optimi- Peherstorfer B, Cui T, Marzouk Y, Willcox K (2016) Multifidel-
zation of aeroelastic systems with structural and aerodynamic ity importance sampling. Comput Methods Appl Mech Eng
uncertainties. J Aircr 50(3):708–715 300:490–509
Nikolaidis E, Chen S, Cudney H, Haftka RT, Rosca R (2004) Compari- Peherstorfer B, Willcox K, Gunzburger M (2018) Survey of mul-
son of probability and possibility for design against catastrophic tifidelity methods in uncertainty propagation, inference, and
failure under uncertainty. J Mech Des 126(3):386–394 optimization. SIAM Rev 60(3):550–591
Nikolaidis E, Ghiocel D, Singhal S (2004) Engineering design reli- Periçaro G, Santos S, Ribeiro A, Matioli L (2015) HLRF-BFGS
ability handbook. CRC Press, Boca Raton optimization algorithm for structural reliability. Appl Math
Noh Y, Choi K, Du L (2009) Reliability-based design optimization Model 39(7):2025–2035
of problems with correlated input variables using a Gaussian Picheny V, Kim N, Haftka R, Queipo N (2008) Conservative predic-
copula. Struct Multidiscip Optim 38(1):1–16 tions using surrogate modeling. In: 49th AIAA/ASME/ASCE/
Noh Y, Choi K, Lee I (2010) Identification of marginal and joint CDFs AHS/ASC Structures, Structural Dynamics, and Materials
using Bayesian method for RBDO. Struct Multidiscip Optim Conference. In: 16th AIAA/ASME/AHS adaptive structures
40(1–6):35 conference, 10th aiaa non-deterministic approaches confer-
Noh Y, Choi K, Lee I, Gorsich D, Lamb D (2011a) Reliability-based ence, 9th AIAA gossamer spacecraft forum, 4th AIAA multi-
design optimization with confidence level for non-Gaussian dis- disciplinary design optimization specialists conference
tributions using bootstrap method. J Mech Des 133(9):091001 Platz R, Götz B (2017) Non-probabilistic uncertainty evaluation
in the concept phase for airplane landing gear design. Model

13
 E. Acar et al.

validation and uncertainty quantification, vol 3. Springer, Santosh T, Saraf R, Ghosh A, Kushwaha H (2006) Optimum step
Cham, pp 161–169 length selection rule in modified HL-RF method for structural
Qiu Z, Wang X (2003) Comparison of dynamic response of struc- reliability. Int J Press Vessels Pip 83(10):742–748
tures with uncertain-but-bounded parameters using non-prob- Schueller G, Pradlwarter H (2007) Benchmark study on reliabil-
abilistic interval analysis method and probabilistic approach. ity estimation in higher dimensions of structural systems—an
Int J Solids Struct 40(20):5423–5439 overview. Struct Saf 29:167–182
Qiu Z, Wang X (2005) Parameter perturbation method for dynamic Schuëller GI, Jensen HA (2008) Computational methods in optimiza-
responses of structures with uncertain-but bounded parameters tion considering uncertainties-an overview. Comput Methods
based on interval analysis. Int J Solids Struct 42(18–19):4970 Appl Mech Eng 198(1):2–13
Qiu Z, Yang D, Elishakoff I (2008) Probabilistic interval reliability Schwarz G (1978) Estimating the dimension of a model. Ann Stat
of structural systems. Int J Solids Struct 45(10):2850–2860 6(2):461–464
Qu X, Haftka R, Venkataraman S, Johnson T (2003) Deterministic Šehić K, Karamehmedović M (2020) Estimation of failure probabili-
and reliability-based optimization of composite laminates for ties via local subset approximations. arxiv:​20030​5994
cryogenic environments. AIAA J 41(10):2029–2036 Shafer G (1976) A mathematical theory of evidence, vol 42. Prince-
Rackwitz R (2001) Reliability analysis-a review and some perspec- ton University Press, Princeton
tives. Struct Saf 23(4):365–395 Shahraki AF, Noorossana R (2014) Reliability-based robust design
Radaideh M, Kozlowski T (2020) Surrogate modeling of advanced optimization: a general methodology using genetic algorithm.
computer simulations using deep Gaussian processes. Reliab Comput Ind Eng 74:199–207
Eng Syst Saf 195:106731 Shan S, Wang G (2008) Reliable design space and complete single-
Rahman S, Wei D (2006) A univariate approximation at most prob- loop reliability-based design optimization. Reliab Eng Syst Saf
able point for higher-order reliability analysis. Int J Solids 93(8):1218–1230
Struct 43(9):2820–2839 Shi L, Lin S (2016) A new RBDO method using adaptive response
Rahman S, Xu H (2004) A univariate dimension-reduction method surface and first-order score function for crashworthiness
for multi-dimensional integration in stochastic mechanics. design. Reliab Eng Syst Saf 156:125–133
Probab Eng Mech 19(4):393–408 Shi Y, Lu Z (2019) Dynamic reliability analysis model for structure
Rajabi M (2019) Review and comparison of two meta-model-based with both random and interval uncertainties. Int J Mech Mater
uncertainty propagation analysis methods in groundwater Des 15(3):521–537
applications: polynomial chaos expansion and Gaussian pro- Shin J, Lee I (2014) Reliability-based vehicle safety assessment and
cess emulation. Stoch Env Res Risk Assess 33(2):607–631 design optimization of roadway radius and speed limit in windy
Rajan A, Luo FJ, Kuang YC, Bai Y, Ooi MPL (2020) Reliability- environments. J Mech Des 136(8):081006
based design optimisation of structural systems using high- Sim J, Qiu Z, Wang X (2007) Modal analysis of structures with uncer-
order analytical moments. Struct Saf 86:101970 tain-but-bounded parameters via interval analysis. J Sound Vib
Ramakrishnan B, Rao S (1996) A general loss function based optimi- 303(1–2):29–45
zation procedure for robust design. Eng Optim 25(4):255–276 Simon C, Bicking F (2017) Hybrid computation of uncertainty in reli-
RAMDO (2021) Reliability analysis, and design optimization soft- ability analysis with p-box and evidential networks. Reliab Eng
ware—-ramdo. https://​www.​altair.​com/​ramdo/ Syst Saf 167:629–638
Ramu P, Qu X, Youn B, Haftka R, Choi K (2006) Inverse reliability Simon D (2002) Sum normal optimization of fuzzy membership func-
measures and reliability-based design optimisation. Int J Reliab tions. Int J Uncertain Fuzz Knowl-Based Syst 10(04):363–384
Saf 1(1–2):187–205 Smarslok B, Haftka R, Carraro L, Ginsbourger D (2010) Improving
Ranjbar A, Mahjouri N (2019) Multi-objective freshwater manage- accuracy of failure probability estimates with separable Monte
ment in coastal aquifers under uncertainty in hydraulic param- Carlo. Int J Reliab Saf 4(4):393–414
eters. Nat Resour Res 29:1–22 SmartUQ (2021) Uncertainty propagation—-smartuq. https://​www.​
Rao S, Berke L (1997) Analysis of uncertain structural systems using smart​uq.​com/​softw​are/​uncer​tainty-​propa​gation/
interval analysis. AIAA J 35(4):727–735 Smirnoff N (1939) Sur les écarts de la courbe de distribution empi-
Rao SS (1992) Reliability-based design. McGraw-Hill Companies, rique. Matematicheskii Sbornik 48(1):3–26
New York Sohouli A, Yildiz M, Suleman A (2018) Efficient strategies for relia-
Romero V, Swiler L, Giunta A (2004) Construction of response bility-based design optimization of variable stiffness composite
surfaces based on progressive-lattice-sampling experimental structures. Struct Multidiscip Optim 57(2):689–704
designs with application to uncertainty propagation. Struct Saf Son H, Lee G, Kang K, Kang Y, Youn B, Lee I, Noh Y (2020) Indus-
26(2):201–219 trial issues and solutions to statistical model improvement: a
Ronold KO, Larsen GC (2000) Reliability-based design of wind- case study of an automobile steering column. Struct Multidiscip
turbine rotor blades against failure in ultimate loading. Eng Optim 61(4):1739–1756
Struct 22(6):565–574 Song J, Kang W (2009) System reliability and sensitivity under statisti-
Rowe W (1994) Understanding uncertainty. Risk Anal 14(5):743–750 cal dependence by matrix-based system reliability method. Struct
Roy CJ, Oberkampf WL (2011) A comprehensive framework for Saf 31(2):148–156
verification, validation, and uncertainty quantification in Soroudi A, Keane A (2015) Risk averse energy hub management con-
scientific computing. Comput Methods Appl Mech Eng sidering plug-in electric vehicles using information gap deci-
200(25–28):2131–2144 sion theory. Plug in electric vehicles in smart grids. Springer,
Sandgren E, Cameron TM (2002) Robust design optimization of Singapore, pp 107–127
structures through consideration of variation. Comput Struct Soroudi A, Rabiee A, Keane A (2017) Information gap decision theory
80(20–21):1605–1613 approach to deal with wind power uncertainty in unit commit-
Sankararaman S, Mahadevan S (2011) Likelihood-based representa- ment. Electr Power Syst Res 145:137–148
tion of epistemic uncertainty due to sparse point data and/or Soundappan P, Nikolaidis E, Haftka R, Grandhi R, Canfield R (2004)
interval data. Reliab Eng Syst Saf 96(7):814–824 Comparison of evidence theory and Bayesian theory for uncer-
tainty modeling. Reliab Eng Syst Saf 85(1–3):295–311

13
Modeling, analysis, and optimization under uncertainties: a review

Sun G, Li G, Zhou S, Li H, Hou S, Li Q (2011) Crashworthiness design estimation. In: 10th AIAA/ISSMO multidisciplinary analysis
of vehicle by using multiobjective robust optimization. Struct and optimization conference, p 4359
Multidiscip Optim 44(1):99–110 Wang L, Wang X, Li Y, Hu J (2019a) A non-probabilistic time-vari-
Sun G, Zhang H, Fang J, Li G, Li Q (2017) Multi-objective and multi- ant reliable control method for structural vibration suppression
case reliability-based design optimization for tailor rolled blank problems with interval uncertainties. Mech Syst Signal Process
(TRB) structures. Struct Multidiscip Optim 55(5):1899–1916 115:301–322
Taflanidis A, Beck J (2008) An efficient framework for optimal robust Wang X, Wang Y (2015a) Nonparametric multivariate density estima-
stochastic system design using stochastic simulation. Comput tion using mixtures. Stat Comput 25(2):349–364
Methods Appl Mech Eng 198(1):88–101 Wang X, Wang L, Elishakoff I, Qiu Z (2011) Probability and convexity
Taflanidis A, Beck J (2008) Stochastic subset optimization for optimal concepts are not antagonistic. Acta Mech 219(1–2):45–64
reliability problems. Probab Eng Mech 23(2–3):324–338 Wang Y (2007) On fast computation of the non-parametric maximum
Tang Y, Chen J, Wei J (2012) A sequential algorithm for reliability- likelihood estimate of a mixing distribution. J R Stat Soc Ser B
based robust design optimization under epistemic uncertainty. J 69(2):185–198
Mech Des 134(1):014502 Wang Z, Chen W (2017) Confidence-based adaptive extreme response
Teckentrup A, Jantsch P, Webster C, Gunzburger M (2015) A multi- surface for time-variant reliability analysis under random excita-
level stochastic collocation method for partial differential equa- tion. Struct Saf 64:76–86
tions with random input data. SIAM/ASA J Uncertain Quantif Wang Z, Wang P (2012) A nested extreme response surface approach
3(1):1046–1074 for time-dependent reliability-based design optimization. J Mech
Thom H (1960) Distributions of extreme winds in the united states. Des 134:121007
Trans Am Soc Civ Eng 126(2):450–462 Wang Z, Wang P (2014) A maximum confidence enhancement based
Toft HS, Sørensen JD (2011) Reliability-based design of wind turbine sequential sampling scheme for simulation-based design. J Mech
blades. Struct Saf 33(6):333–342 Des 136(2):021006
Tonon F, Bernardini A, Elishakoff I (2001) Hybrid analysis of uncer- Wang Z, Wang P (2015b) An integrated performance measure approach
tainty: probability, fuzziness and anti-optimization. Chaos Soli- for system reliability analysis. J Mech Des 137(2):021406
tons Fract 12(8):1403–1414 Wang Z, Wang Z, Yu S, Zhang K (2019b) Time-dependent mechanism
Tripathy R, Bilionis I (2018) Deep UQ: learning deep neural network reliability analysis based on envelope function and vine-copula
surrogate models for high dimensional uncertainty quantification. function. Mech Mach Theory 134:667–684
J Comput Phys 375:565–588 Wang Z, Li H, Chen Z, Li L, Hong H (2020) Sequential optimiza-
Tripathy R, Bilionis I, Gonzalez M (2016) Gaussian processes with tion and moment-based method for efficient probabilistic design.
built-in dimensionality reduction: applications to high-dimen- Struct Multidiscip Optim 62:1–18
sional uncertainty propagation. J Comput Phys 321:191–223 Weinmeister J, Xie N, Gao X, Krishna Prasad A, Roy S (2018) Analy-
Tu J, Choi K, Park Y (1999) A new study on reliability-based design sis of a polynomial chaos-Kriging metamodel for uncertainty
optimization. J Mech Des 121(4):557 quantification in aerospace applications. In: 2018 AIAA/ASCE/
Tu J, Choi K, Park Y (2001) Design potential method for robust system AHS/ASC structures, structural dynamics, and materials confer-
parameter design. AIAA J 39(4):667–677 ence, p 0911
UQWorld (2021) Various uncertainty quantification software tools. Wu X, Mui T, Hu G, Meidani H, Kozlowski T (2017) Inverse uncer-
https://​uqwor​ld.​org/t/​vario​us-​uncer ​tainty-​quant​ifica​tion-​softw​ tainty quantification of trace physical model parameters using
are-​tools/​137/ sparse gird stochastic collocation surrogate model. Nucl Eng Des
Valdebenito M, Schuëller G (2010) A survey on approaches for reliabil- 319:185–200
ity-based optimization. Struct Multidiscip Optim 42(5):645–663 Wu X, Kozlowski T, Meidani H (2018) Kriging-based inverse uncer-
Viana F, Haftka R, Steffen V (2009) Multiple surrogates: how cross- tainty quantification of nuclear fuel performance code bison
validation errors can help us to obtain the best predictor. Struct fission gas release model using time series measurement data.
Multidiscip Optim 39(4):439–457 Reliab Eng Syst Saf 169:422–436
Viana F, Picheny V, Haftka R (2010) Using cross validation to design Wu Y, Y S, Sues R, Cesare M (2001) Safety factor based approach for
conservative surrogates. AIAA J 48(10):2286–2298 probability–based design optimization. In: Proceedings of 42nd
Volpi S, Diez M, Gaul N, Song H, Iemma U, Choi K, Stern F (2015) AIAA/ASME/ASCE/AHS/ASC structures, structural dynamics
Development and validation of a dynamic metamodel based on and materials conference, Seattle, WA
stochastic radial basis functions and uncertainty quantification. Wunsch D, Hirsch C, Nigro R, Coussement G (2015) Quantifica-
Struct Multidiscip Optim 51(2):347–368 tion of combined operational and geometrical uncertainties in
Wand M, Jones M (1994) Multivariate plug-in bandwidth selection. turbo-machinery design. In: Turbo expo: power for land, sea,
Comput Stat 9(2):97–116 and air. American Society of Mechanical Engineers, vol 56659,
Wang C, Matthies H (2019) Novel model calibration method via non- p V02CT45A018
probabilistic interval characterization and Bayesian theory. Xi Z (2019) Model-based reliability analysis with both model uncer-
Reliab Eng Syst Saf 183:84–92 tainty and parameter uncertainty. J Mech Des 141(5):051404
Wang C, Duan Q, Tong CH, Di Z, Gong W (2016) A gui platform Xiao M, Zhang J, Gao L (2020) A system active learning Kriging
for uncertainty quantification of complex dynamical models. method for system reliability-based design optimization with a
Environ Modell Softw 76:1–12. https://​doi.​org/​10.​1016/j.​envso​ multiple response model. Reliab Eng Syst Saf 199:106935
ft.​2015.​11.​004 Xiao Z, Han X, Jiang C (2016) An efficient uncertainty propaga-
Wang C, Zhang H, Beer M (2018) Computing tight bounds of structural tion method for parameterized probability boxes. Acta Mech
reliability under imprecise probabilistic information. Comput 227:633–649
Struct 208:92–104 Xiong Y, Chen W, Tsui K, Apley D (2009) A better understanding
Wang F, Xiong F, Chen S, Song J (2019) Multi-fidelity uncertainty of model updating strategies in validating engineering models.
propagation using polynomial chaos and Gaussian process mod- Comput Methods Appl Mech Eng 198(15–16):1327–1337
eling. Struct Multidiscip Optim 60(4):1583–1604 Xu H, Rahman S (2004) A generalized dimension-reduction method
Wang L, Beeson D, Wiggs G (2004) Efficient and accurate point for multidimensional integration in stochastic mechanics. Int J
estimate method for moments and probability distribution Numer Meth Eng 61(12):1992–2019

13
 E. Acar et al.

Xu J, Wang D (2019) Structural reliability analysis based on polyno- Zhang D, Han X, Jiang C, Liu J, Li Q (2017) Time-dependent reli-
mial chaos, Voronoi cells and dimension reduction technique. ability analysis through response surface method. J Mech Des
Reliab Eng Syst Saf 185:329–340 139:041404
Yadav OP, Bhamare SS, Rathore A (2010) Reliability-based robust Zhang H, Mullen R, Muhanna R (2010a) Finite element structural
design optimization: a multi-objective framework using hybrid analysis using imprecise probabilities based on p-box represen-
quality loss function. Qual Reliab Eng Int 26(1):27–41 tation. In: The 4th international workshop on reliable engineering
Yang D (2010) Chaos control for numerical instability of first order computing. Professional Activities Centre, National University
reliability method. Commun Nonlinear Sci Numer Simul of Singapore
15(10):3131–3141 Zhang H, Mullen R, Muhanna R (2010b) Interval Monte Carlo methods
Yang M, Zhang D, Han X (2020) New efficient and robust method for for structural reliability. Struct Saf 32(3):183–190
structural reliability analysis and its application in reliability- Zhang H, Mullen R, Muhanna R (2011) Structural analysis with prob-
based design optimization. Comput Methods Appl Mech Eng ability-boxes. Int J Reliab Saf 6(1–3):110–129
366:113018 Zhang J (2011) Adaptive normal reference bandwidth based on quan-
Yang R, Gu L (2004) Experience with approximate reliability- tile for kernel density estimation. J Appl Stat 38(12):2869–2880
based optimization methods. Struct Multidiscip Optim Zhang J, Du X (2010) A second-order reliability method with first-
26(1–2):152–159 order efficiency. J Mech Des 132(10):101006
Yang X, Liu Y, Mi C, Wang X (2018) Active learning Kriging model Zhang J, Taflanidis A (2019) Multi-objective optimization for design
combining with kernel-density-estimation-based importance under uncertainty problems through surrogate modeling in aug-
sampling method for the estimation of low failure probability. mented input space. Struct Multidiscip Optim 59(2):351–372
J Mech Des 140:051402 Zhang X, King M, Hyndman R (2006) A Bayesian approach to band-
Yoo D, Lee I (2014) Sampling-based approach for design optimization width selection for multivariate kernel density estimation. Com-
in the presence of interval variables. Struct Multidiscip Optim put Stat Data Anal 50(11):3009–3031
49(2):253–266 Zhang X, Wang L, Sørensen J (2020) AKOIS: an adaptive Kriging
Yoo D, Lee I, Cho H (2014) Probabilistic sensitivity analysis for novel oriented importance sampling method for structural system reli-
second-order reliability method (SORM) using generalized chi- ability analysis. Struct Saf 82:10876
squared distribution. Struct Multidiscip Optim 50(5):787–797 Zhang Z, Wang J, Jiang C, Huang Z (2019) A new uncertainty propa-
Youn B, Choi K (2004) An investigation of nonlinearity of reli- gation method considering multimodal probability density func-
ability based design optimization approaches. J Mech Des tions. Struct Multidiscip Optim 60(5):1983–1999
126(3):403–411 Zhao L, Choi K, Lee I, Gorsich D (2013) Conservative surrogate model
Youn B, Wang P (2008) Bayesian reliability-based design optimization using weighted Kriging variance for sampling-based RBDO. J
using eigenvector dimension reduction (EDR) method. Struct Mech Des 135(9):091003
Multidiscip Optim 36(2):107–123 Zheng Y, Qiu Z (2018) Non-probabilistic stability reliability analysis
Youn B, Choi K, Park Y (2003) Hybrid analysis method for reliability- of composite laminated panels in supersonic flow with uncer-
based design optimization. J Mech Des 125(2):221 tain-but-bounded parameters. In: 2018 AIAA non-deterministic
Youn B, Choi K, Yang R, Gu L (2004) Reliability-based design optimi- approaches conference, p 0438
zation for crashworthiness of vehicle side impact. Struct Multi- Zhou T, Peng Y (2020) Structural reliability analysis via dimension
discip Optim 26:272–283 reduction, adaptive sampling, and Monte Carlo simulation. Struct
Youn B, Choi K, Du L (2005a) Adaptive probability analysis using an Multidiscip Optim 62(5):2629–2651
enhanced hybrid mean value method. Struct Multidiscip Optim Zhou XY, Ruan X, Gosling P (2019a) Robust design optimization of
29(2):134–148 variable angle tow composite plates for maximum buckling load
Youn BD, Xi Z (2009) Reliability-based robust design optimization in the presence of uncertainties. Compos Struct 223:110985
using the eigenvector dimension reduction (edr) method. Struct Zhou Y, Lu Z (2019) Active polynomial chaos expansion for reliability-
Multidiscip Optim 37(5):475–492 based design optimization. AIAA J 57(12):5431–5446
Youn BD, Choi KK, Yi K (2005b) Performance moment integra- Zhou Y, Lu Z, Cheng K (2019b) Sparse polynomial chaos expansions
tion (pmi) method for quality assessment in reliability-based for global sensitivity analysis with partial least squares and dis-
robust design optimization. Mech Based Des Struct Mach tance correlation. Struct Multidiscip Optim 59(1):229–247
33(2):185–213 Zhu P, Shi L, Yang R, Lin S (2015) A new sampling-based RBDO
Youn BD, Choi KK, Du L, Gorsich D (2007) Integration of possibility- method via score function with reweighting scheme and appli-
based optimization and robust design for epistemic uncertainty cation to vehicle designs. Appl Math Model 39(15):4243–4256
Zadeh L (1965) Fuzzy sets. J Inf Control 8:338–353 Zhu Z, Du X (2016) Reliability analysis with Monte Carlo simulation
Zadeh L (1973) Outline of a new approach to the analysis of complex and dependent Kriging predictions. J Mech Des 138(12):121403
systems and decision processes. IEEE Trans Syst Man Cybern Zimmermann H (2001) Fuzzy analysis. In: Fuzzy set theory and its
1:28–44 applications. Springer, Dordrecht
Zadeh L (1978) Fuzzy sets as a basis for a theory of possibility. Fuzzy Zio E, Pedroni N (2013) Literature review of methods for representing
Sets Syst 1(1):3–28 uncertainty. FonCSI
Zafar T, Wang Z (2020) Time-dependent reliability prediction using Zou T, Mahadevan S (2006) A direct decoupling approach for efficient
transfer learning. Struct Multidiscip Optim 62:147–158 reliability-based design optimization. Struct Multidiscip Optim
Zaman K, Mahadevan S (2013) Robustness-based design optimization 31(3):190
of multidisciplinary system under epistemic uncertainty. AIAA Zougab N, Adjabi S, Kokonendji C (2014) Bayesian estimation of
J 51(5):1021–1031 adaptive bandwidth matrices in multivariate kernel density esti-
Zaman K, Mahadevan S (2017) Reliability-based design optimization mation. Comput Stat Data Anal 75:28–38
of multidisciplinary system under aleatory and epistemic uncer- Zuev K, Beck J, Au S, Katafygiotis L (2012) Bayesian post-processor
tainty. Struct Multidiscip Optim 55(2):681–699 and other enhancements of subset simulation for estimating fail-
Zang C, Friswell M, Mottershead J (2005) A review of robust opti- ure probabilities in high dimensions. Comput Struct 92:283–296
mal design and its application in dynamics. Comput Struct
83(4–5):315–326

13
Modeling, analysis, and optimization under uncertainties: a review

Publisher's Note Springer Nature remains neutral with regard to

jurisdictional claims in published maps and institutional affiliations.

13