Chinese Journal of Aeronautics, (2018), 31(1): 41–53

Chinese Society of Aeronautics and Astronautics

& Beihang University
Chinese Journal of Aeronautics
cja@buaa.edu.cn
www.sciencedirect.com

A Bayesian approach for integrating multilevel

priors and data for aerospace system reliability
assessment
Jian GUO, Zhaojun LI *, Thomas KEYSER

Department of Industrial Engineering and Engineering Management, Western New England University, Springfield, MA 01119, USA

Received 6 December 2016; revised 21 December 2016; accepted 7 March 2017

Available online 6 December 2017

KEYWORDS Abstract This paper investigates Bayesian methods for aerospace system reliability analysis using
Bayesian inference; various sources of test data and expert knowledge at both subsystem and system levels. Four sce-
Bayesian melding; narios based on available information for the priors and test data of a system and/or subsystems are
Multilevel information; studied using specific Bayesian inference techniques. This paper proposes the Bayesian melding
Markov Chain Monte Carlo method for integrating subsystem-level priors with system-level priors for both system- and
(MCMC); subsystem-level reliability analysis. System and subsystem reliability outcomes are compared under
Sampling Importance Re- different scenarios. Computational challenges for posterior inferences using the sophisticated Baye-
sampling (SIR); sian melding method are addressed using Markov Chain Monte Carlo (MCMC) and adaptive Sam-
System reliability pling Importance Re-sampling (SIR) methods. A case study with simulation results illustrates the
applications of the proposed methods and provides insights for aerospace system reliability analysis
using available multilevel information.
Ó 2017 Chinese Society of Aeronautics and Astronautics. Production and hosting by Elsevier Ltd. This is
an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

1. Introduction and other factors which have impacts on product reliability.

In the product development process for aerospace systems,
system-level reliability testing is very costly and can end up
with very limited test data. Product experts may have valuable
reliability knowledge based on understanding of adopted new
technologies, complexity of a new product, manufacturability,
* Corresponding author.
E-mail address: zhaojun.li@wne.edu (Z. LI).
Peer review under responsibility of Editorial Committee of CJA.
Production and hosting by Elsevier
Integrating limited testing data with expert reliability knowl-
edge can significantly improve aerospace system reliability esti-
mation with significant potential cost savings on system-level
reliability testing. When there exist different levels of reliability
prior information and testing data at system and/or subsystem
levels, integrating such multilevel information can be challeng-
ing. Bayesian inference is one of the most viable methods to
quantify expert reliability knowledge through appropriate
prior distribution elicitation. In addition, Bayesian methods
are capable of dealing with limited aerospace testing data.1
Recently, Bayesian reliability inference has received increasing
applications and acceptance for reliability estimation when
there exists limited data.2

https://doi.org/10.1016/j.cja.2017.08.020
1000-9361 Ó 2017 Chinese Society of Aeronautics and Astronautics. Production and hosting by Elsevier Ltd.
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
42
Nomenclature
h unknown parameter(s) in a probability distribu-
tion function
pe expert’s estimation on h
Cp confidence level of the expert’s estimation on h
pðÞ posterior probability distribution
pðÞ posterior probability distribution
qðÞ proposal probability distribution in Metropolis-
Hasting algorithm
q ðÞ induced prior distribution of the system reliability
parameter
q~ðÞ pooled prior distribution
For a complex aerospace system, system- or higher-level
module reliability is dependent on subsystem- or lower-level
module reliability performance. Thus, system reliability can
be assessed based on lower-level module reliability perfor-
mance. Inversely, system reliability information can be used
for subsystem- and module-level reliability inferences and
updates. In this research, we investigate four common scenar-
ios based on the availability of different levels of prior reliabil-
ity information and data shown in Fig. 1. These four scenarios
are (A) using the subsystem data and prior; (B) using the sub-
system prior and data as well as the system prior; (C) using the
subsystem prior as well as the system prior and data; and (D)
using both system and subsystem data and priors. To obtain
the posterior system reliability inference for each of these four
scenarios, specific Bayesian methods are suggested in order to
better deal with the unique data and priors available.
When only a single level of data and prior information is
available, Bayesian reliability inference for a unit, e.g., a com-
ponent, subsystem, or system, can proceed by following the
standard Bayesian procedure.1,2 For system reliability analysis
using the prior and data of components/subsystems, extensive
research has been conducted under the Bayesian inference
framework. Springer and Thompson3,4 developed the Bayesian
reliability confidence intervals of a product with N indepen-
dent subsystems using the beta prior and binary test data.
Fig. 1 Various scenarios for aerospace system and subsystem reliability assessment.
J. GUO et al.
qI ðÞ
probability distribution of input parameters
qO ðÞ
probability distribution of output parameters
MðÞ deterministic function mapping two-parameter
vectors
JðÞ Jacobian matrix
I set of model input parameters
O set of model output parameters
MCMC Markov Chain Monte Carlo
MH Metropolis-Hastings
SIR Sampling Importance Re-sampling
Tang et al.5 proposed a Bayesian method along with the
moment function to estimate system reliability with compo-
nent binary test data and Beta priors for series, parallel, and
mixed systems. Singh et al.6 applied the Bayesian approach
in component-based software development by including a sys-
tem’s internal structure. Assuming exponentially-distributed
failure time, Mastran7 and Mastran and Singpurwalla8 applied
Bayesian methods for reliability assessment of a coherent sys-
tem using binary test data and priors from both components
and the system. Martz et al.9,10 studied the overall reliability
of an air-to-air heat-seeking missile system using a two-stage
Bayesian procedure.
To aggregate the multilevel prior and data from both a sys-
tem and its subsystems, a Natural-Conjugate (NC) method
proposed by Winkler11 was used to average the induced prior
and the native prior, in which the induced prior is derived
based on the system structure and the native prior is the exist-
ing system-level prior. In addition, prior knowledge can be
incorporated through a hierarchical model where the parame-
ters of reliability distributions are assigned to follow certain
distributions or dynamic functions to resolve estimation aggre-
gation. Hamada et al.12 combined multilevel failure informa-
tion in a fault tree by multiplying the high-level event
probability in the posterior distribution, which assigned equal
weights to the native and induced beta prior distributions.
A Bayesian approach for integrating multilevel priors and data
Bier13 systematically discussed the consistency between the
results of aggregated system-level reliability and disaggregated
component-level reliability analysis of the same system using
Bayesian estimation, and observed the two avenues rarely con-
cur for complex systems. In order to address the consistency
issue in Ref.13, Johnson et al.14 proposed a Bayesian hierarchi-
cal model to incorporate reliability point estimation from
experts with binary test data. Wilson et al.15 summarized Baye-
sian hierarchical methods for binary, lifetime, and degradation
data. The Bayesian melding method, which was initially pro-
posed by Poole and Raftery16 to evaluate the population evo-
lution of bow-head whales, has also been applied to integrate
subsystem-level priors with system priors for system posterior
reliability inference. For example, Reese et al.17 and Guo and
Wilson18 studied how to integrate heterogeneous multilevel
information in assessment of system reliability using the Baye-
sian melding method. However, very little literature exists per-
taining to using system-level reliability information to update
subsystem-level posterior reliability under various scenarios
in terms of availability of multilevel prior and data. This is a
unique aspect of this research.
Computational challenges of Bayesian inference are often
related to complex integrations for computing and evaluating
posterior distributions, marginal distributions, and posterior-
based predictions. For cases with non-conjugate priors and
high-dimensional parameter space, numerical methods such
as Approximate Bayesian Computation (ABC),19 importance
sampling, iterative quadrature, Laplace approximation, Monte
Carlo Markov Chain (MCMC), and variational Bayes have
been used. For many high-dimensional problems, MCMC
methods could be the only choices.20 MCMC methods are a
practical set of computational tools which exploit the proba-
bilistic properties of integrands rather than the analytic prop-
erties.21 In Bayesian reliability analysis, MCMC methods are
capable of dealing with complex parametrization and likeli-
hood functions. Specifically, Gibbs sampling22 and
Metropolis-Hastings (MH)23,24 algorithms are the two most
popular computational methods for Bayesian inference. Gibbs
sampling is an iterative sampling technique overcoming the
dimensionality issue when full conditionals are available.25
When parameter spaces have different dimensionalities and
conditional distributions are not in closed forms, the MH algo-
rithm, a more generic version of Gibbs sampling, becomes nec-
essary. However, for the proposed Bayesian system reliability
analysis using the Bayesian melding method in which compo-
nent and system priors as well as the induced system prior have
been sampled, the traditional MCMC method becomes ineffi-
cient due to repeated likelihood function evaluations in the
MCMC sampling process. Such issues become more serious
when a large number of sampling data is needed and the com-
plicated posterior distribution consists of multiple-level priors
and complex likelihood functions. Thus, the Sampling Impor-
tance Re-sampling (SIR) method with less algorithm complex-
ity is suggested in the proposed Bayesian melding method for
aerospace systems reliability assessment.
The remainder of the paper is organized as follows. Sec-
tion 2 presents the basic procedure of Bayesian methods in reli-
ability applications with an overview of the three key elements
in Bayesian statistics inference, i.e., prior elicitation pðhÞ, sam-
pling model yðhÞ, and posterior inference pðhÞ. Section 3 inves-
tigates different ways of using Bayesian methods for assessing
system reliability based on the availability of system and
43
subsystem prior distributions and data (Fig. 1). An aerospace
system is presented as a case study to demonstrate the applica-
tions of the proposed methods in Section 4. Section 5 draws
conclusions and discusses future research work.
2. Bayesian procedure for reliability assessment
The three key elements in Bayesian statistics inference are prior
elicitation, sampling model, and posterior computation. Prior
information reflects existing knowledge about the unknown
parameter h before any data is collected. Sampling models
describe the frequency variation over the observed data,
through which the test data is described in likelihood func-
tions. The posterior distribution can then be evaluated through
integrating the prior information and the likelihood functions.
When complex priors and sampling models are used, determin-
istic closed forms of posterior distributions usually do not
exist. In such cases, simulation techniques such as MCMC
can be used to assess posterior distributions and proceed with
further Bayesian inference and Bayesian predictive inferences.
Simulations are also used to resolve computational difficulties
in high-dimensional parameterization cases to avoid the ‘‘curse
of dimensionality”.
2.1. Prior elicitation, sampling model, and posterior inference
Bayesian inference often starts with eliciting a prior distribu-
tion to appropriately model existing knowledge, previous
experiences, and subjective beliefs through a probability state-
ment about the unknown parameter h. In reliability engineer-
ing, expert engineering judgment can provide important
references for modeling the reliability of a system.26 Expert
judgment should follow the law of probability,27 i.e., estima-
tions on the key parameters of a prior distribution need to
come with a corresponding probability. Expert engineers
can either provide parameter estimations of prior distribu-
tions of interest directly or present the estimates of certain
key parameters in the prior such as a lower percentile value
and an associated confidence level. When conjugate prior dis-
tributions are applicable, the posterior distributions are
within the same family of probability distributions. Hamada
et al.2 summarized various types of reliability data, sampling
distributions, and prior distributions. For binary data (pass/-
fail), the binomial sampling model with its conjugate beta dis-
tribution has been widely used. Weiler28 investigated methods
of estimating the parameters of a beta prior for various types
of historical observations including the expectation of the rate
EðhÞ, the probability of Pðh > h0 Þ, and the numbers of suc-
cess events in past N trials. Specifically, a beta distribution
is defined as
Cða þ bÞ a1
pðhja; bÞ ¼ h ð1  hÞb1 ð1Þ
CðaÞCðbÞ
where h, the parameter of interest, is the probability that an
experiment passes the pass/fail test and CðÞ is the gamma
function. Parameters a and b are the shape parameters of the
beta distribution, which are usually called super parameters
in Bayesian inferences. These super parameters can be deter-
mined based on expert estimation about h, for example, the
estimated mode or mean of h and a corresponding confidence
44
level of the mode or mean estimate based on historical pass/fail
test outcomes.
In addition to using point estimates for h, an interval esti-
mate for h with a confidence level can also be used to deter-
mine the prior distribution.29 The confidence level or weight
can be defined in a relative scale using natural numbers. For
instance, a value of confidence level of 20 indicates a high con-
fidence level, while a confidence level of 5 indicates a low con-
fidence level for the associated estimates, e.g., the mode or
mean for h. Denote expert estimation on the mode of h and
the confidence level as pe and Cp , respectively. Then a and b
can be parameterized as Cp pe þ 1 and Cp ð1  pe Þ þ 1,
respectively.14
When other types of test data exist, probability models such
as Poisson, exponential, and Weibull models can also be used
for prior distribution modeling. Various types of prior and
posterior distributions are suggested in the literature.2 Taking
the Poisson distribution with counting failure data as an exam-
ple, parameter k. rival rate k in Poisson distribution
Y  Poisson kt models failure counting data at a time point
of t, where Y is the total number of occurrences. The conjugate
gamma prior distribution Gammaða; bÞ with the shape param-
eter a and the scale parameter b for the Poisson sampling
model can be employed to obtain a closed-form posterior dis-
tribution for parameter k.
Posterior distribution inference integrates prior informa-
tion and observed data through a likelihood function. With
a conjugate prior distribution, the posterior distribution can
be easily evaluated. Using the previous parameterization nota-
tions pe and Cp for a beta prior distribution, the posterior dis-
tribution with binary test data is
CðCp þ 2 þ nÞ
pðhÞ ¼
CðCp pe þ 1 þ sÞCðCp ð1  pe Þ þ 1 þ n  sÞhCp pe þs ð1  hÞCp ð1pe Þþns
ð2Þ
where s and n are the numbers of passes and total tests,
respectively.
2.2. Bayesian posterior inference using simulation
Applications of Bayesian inference in aerospace system relia-
bility analysis aim to estimate unknown reliability parameters,
determine marginal effects, and predict future failures or suc-
cess probabilities. Such Bayesian statistical inferences can be
computationally intensive due to the non-conjugate prior spec-
ification and high-dimensional parameterization. Often simu-
lation techniques will become imperative. Among various
simulation techniques, MCMC methods have been widely used
to evaluate the posterior reliability of complex systems. Evans
and Swartz20 believed that MCMC methods are the only
choices for high-dimensional problems. MCMC methods aim
to evaluate essential statistics such as the mean, percentiles,
and higher-order moments of the posterior distribution, which
usually does not have a closed-form solution. Such approxima-
tion and simulation methods of evaluating the posterior distri-
bution can be realized through generating a sequence of
samples fh1 ; h2 ; . . . ; hs g so that the following equation holds
under the ergodic theory with a large number of simulation
runs:
J. GUO et al.
Z
1X s
gðhi Þ  gðhÞpðhÞdh ð3Þ
s j¼1
where gðhÞ is any function of interest of the unknown parame-
ter h, which could be a high-dimensional vector. The target dis-
tribution function pðhÞ is the posterior distribution in Bayesian
inference. Liang et al.25 summarized various MCMC algo-
rithms, among which Gibbs sampling22 and Metropolis-
Hastings23,24 algorithms are the most commonly utilized for
Bayesian inference. In order to obtain the posterior distribution
of the unknown parameter h, the Metropolis-Hastings method
proposes a new value h around the current value hs using a
proposal distribution and determines whether to assign h as
the next value of hsþ1 through an acceptance ratio r as follows:
 
h with probability r
hsþ1 ¼ ð4Þ
hs with probability 1  r
pðh Þ qðh ; hs Þ
where r ¼  . The target distribution pðÞ is the
pðhs Þ qðhs ; h Þ
posterior distribution in Bayesian inference, and qðÞ is the pro-
posal distribution. The acceptance decision can be accom-
plished by generating a random number u from uniformð0; 1Þ
and setting hsþ1 ¼ h if r > u, and hsþ1 ¼ hs , otherwise.
The proposal distribution qðÞ from which new samples are
generated is critical for rapid convergence of the Metropolis-
Hastings (MH) algorithm. If pðÞ can be sampled directly,
the fastest-converging proposal distribution would be the tar-
get distribution,30 where the acceptance ratio equals one. Chib
and Greenberg31 proposed a solution to specify the proposal
distribution for a given form of pðÞ. If the target distribution
pðÞ can be written as pðÞ / uðÞhðÞ, where hðÞ can be sam-
pled and uðÞ is uniformly bounded, it is feasible to set hðÞ
as the proposal distribution. Since the posterior distribution
is the product of the prior distribution and a sampling distribu-
tion which is always less than 1, the prior distribution pðÞ
becomes a good candidate for the proposal distribution. The
method of using the prior distribution as the proposal distribu-
tion in the MH algorithm is investigated in this research.
3. System reliability assessment using multilevel information
System reliability depends on reliability performances of sub-
systems as well as complex interactions within a given system
structure. This section investigates different ways of integrat-
ing multilevel data and priors using specific Bayesian inference
techniques.
3.1. System reliability assessment based on a single level of prior
and data
When only a single level of subsystem priors and data is avail-
able, system reliability can be evaluated using the ‘‘bottom-
top” approach based on the system structure. Fig. 2 shows
an example of a mixed system consisting of both parallel and
series subsystems. The success rate or reliability of an individ-
ual subsystem is h under the Bernoulli experimentation. The
system reliability can be expressed as follows:
" #
Yg Y
h Y
Lj
hsys ¼ hi  1  ð1  hk Þ ð5Þ
i¼1 j¼1 k¼1
A Bayesian approach for integrating multilevel priors and data
Fig. 2 Example of a mixed system with parallel and series subsystems.
where g indicates the number of subsystems connected in ser-
ies, h indicates the number of subsystems connected in parallel,
Lj is the number of sub-subsystems/components in the j th par-
allel subsystem, hi is the reliability of the ith subsystem in ser-
ies, and hk is the reliability of the kth sub-subsystem in the jth
parallel subsystem.
Using the mixed system in Fig. 2, with the binary test data
and beta distribution assumptions, the posterior distributions
of h for each subsystem can be evaluated by integrating
subsystem-level priors and test data information. The
system-level posterior can then be evaluated using the reliabil-
ity block diagram analysis and system structure. When the
prior elicitation is realized using expert inputs for Cpi and
pei , the closed-form subsystem posterior reliability can be eval-
uated using Eq. (2). Since there is no closed-form probability
distribution for the system-level posterior distribution, Monte
Carlo (MC) simulations at the subsystem level can be aggre-
gated for system-level posterior reliability inferences. When
non-conjugate priors are used, no closed-form posterior prob-
ability distributions exist for subsystem-level reliability, and
MCMC becomes a viable method for subsystem and system
posterior reliability inferences.
3.2. System reliability assessment based on multiple levels of
prior and data
System reliability assessment based on the system structure can
effectively deal with the situation when only subsystem data
and priors are available. When multilevel data and priors from
subsystems and the system are available, Bayesian melding
methods for integrating such multilevel information are inves-
tigated in this section.
3.2.1. Bayesian melding method for integrating multilevel priors
In the literature, Martz et al.9,10 and Hamada et al.12 inte-
grated multilevel prior information by assigning equal weights
to the native and induced system priors. However, using sys-
tem reliability information to update subsystem-level posterior
distributions has not been well investigated. Instead of averag-
ing multilevel prior distributions using equal weights, the
Bayesian melding method derives the induced prior distribu-
tion via a deterministic model and integrates the induced prior
with the native prior using mathematical pooling algorithms.
The Bayesian melding method was developed based on
Bayesian synthesis32 that is used to simulate the population
evolution process of bowhead whales. In Bayesian synthesis,
the marginal posterior distribution of input parameters is
defined as pI ðIÞ ¼ pIjO ðIjO ¼ MðIÞÞ, where MðÞ is a determin-
istic function, and I and O stand for sets of input and output
parameters, respectively. However, Wolpert33 proved that
Bayesian synthesis has the Borel paradox problem since poste-
rior distribution simulation depends on the selection of the
45
deterministic function MðÞ. The Bayesian melding method
eliminates this deficiency by reformulating via a standard
Bayesian procedure. The induced prior distribution of the out-
put, denoted by qI ðOÞ, integrates information from input
probability distributions qI ðIÞ and the deterministic function
MðIÞ, which means that qI ðOÞ ¼ qI ðM1 ðOÞÞjJðOÞj, given
MðÞ is invertible. The pooled distribution of the output,
q~O ðÞ, can be derived by pooling the induced prior, qI ðÞ, and
the native prior, qO ðÞ. The two main pooling methodologies
are the linear pooling and logarithm pooling methods, which
were initially used to integrate multiple expert opinions. Gen-
est et al.34 proved that the logarithmic pooling operator is the
only pooling operator that can meet the requirement of exter-
nal Bayes property, i.e., leading to the same posterior probabil-
ity distribution regardless of the order of updating priors first
and then combining them, or combining priors first and then
updating them. The pooled prior distribution of the output
under the logarithm pooling algorithm is
q~O ðOÞ / qI ðOÞa qO ðOÞ1a ð6Þ
An updated subsystem prior can be evaluated based on
q~O ðOÞ and MðÞ, i.e., q~O ðOÞ is converted to I space through
O ¼ MðIÞ. Moreover, an updated system prior reliability can
also be obtained from the updated subsystem priors through
the system configuration. Assume that C1 is a hypercube in a
higher-dimensional space and C2 is a hypercube in a lower-
dimensional space mapped from C1 through a convertible
deterministic function MðÞ. C3 is a hypercube in the same
dimensional space as that of C2 via C3 ¼ M1 ðC2 Þ. C1 , C2 ,
and C3 are shown in Fig. 3. In addition, with the assumption
that I 2 C1 , O ¼ MðIÞ 2 C2 , and PðC2 Þ ¼ PðC3 Þ, where PðÞ
denotes any probability distribution. Due to the definitions
of C2 and C3 , qI ðC3 Þ ¼ qI ðC2 Þ. Eqs. (7) and (8) show the
derivation of the pooled distribution of the input, where the
inverse function, M1 ðÞ, is not involved. Thus Eq. (8) also
applies to cases when the deterministic function MðÞ is non-
convertible. It is observed from Eq. (8) that the pooled input
prior distribution q~I ðIÞ is the original prior weighted by the
ratio of two probability densities of qO ðMðIÞÞ and qI ðMðIÞÞ,
evaluated at MðIÞ.
   
qI ðC1 Þ q ðC1 Þ
q~I ðC1 Þ ¼ q~I ðC3 Þ ¼ q~I ðC2 Þ I ð7Þ
qI ðC3 Þ qI ðC2 Þ
Fig. 3 Schematic for prior integration and allocation using
Bayesian melding.
46
 
q ðIÞ
q~I ðIÞ ¼ q~I ðMðIÞÞ  I
qI ðMðIÞÞ
   1a
q ðIÞ q ðMðIÞÞ
/ qI ðOÞa qO ðOÞ1a  I / qI ðIÞ O
qI ðMðIÞÞ qI ðMðIÞÞ
In reliability engineering, system reliability can be referred
as the output of a deterministic function whose input parame-
ters are the reliability of subsystems in h. Based on the mixed
system structure in Fig. 2, the deterministic function is
Qg Q h QLj i (4) The posterior distribution of h is the product of the prior
hsys ¼ i¼1 hi  hj¼1 1  k¼1 ð1  hk Þ , where h is the vector
of hi , the probability/reliability that subsystem i works. The
prior reliabilities qI ðÞ and qO ðÞ are the prior reliabilities of
subsystems and the system, respectively.
3.2.2. Simulation algorithms for posterior inference under
Bayesian melding
According to the standard Bayesian inference, the posterior
can be made inferred with the prior and corresponding data.
Induced and pooled system priors are often very complex
and do not have a closed form in most cases. As a result,
simulation methods such as MCMC methods and impor-
tance sampling methods need to be deployed to obtain their
posteriors. Reese et al.17 and Guo and Wilson18 applied
MCMC in Bayesian melding computation. Taddy et al.35
recommended Sampling Importance Re-sampling (SIR),
which was initially developed by Rubin36 for generating sam-
ples of induced and pooled prior distributions. The SIR
method can simulate and evaluate both subsystem and sys-
tem posteriors more efficiently using samples from subsystem
priors, induced system prior, and pooled system prior distri-
butions. The computation complexity of MCMC methods in
this case is Oðn lg nÞ37 which is larger than OðnÞ of using the
SIR method, where OðÞ describes the limiting behavior of a
function when the argument tends towards a particular value
or infinity. Formally the SIR standard procedure can be
described as follows:
(1) Generate h from qI ðhÞ and evaluate hsys ¼ MðhÞ to
obtain hsys . Assume that subsystems in a complex system
are independent and all follow beta probability distribu-
tions. The joint probability distribution of the subsystem
reliability is:
n 
Y 
Cðai þ bi Þ bi 1
qI ðhÞ ¼ hai i 1 ð1  hi Þ ; as one whose weight is greater than tw ;
i¼1
Cðai ÞCðbi Þ
h ¼ ðh1 ; h2 ; . . . ; hn ÞT
where n is the number of subsystems in the system. ai and bi
are the super parameters of beta distribution of unknown
parameter hi of each subsystem i, which can be evaluated
as Cpi  pei and Cpi  ð1  pei Þ, respectively, according to
Eq. (2).
(2) Evaluate qI ðhsys Þ using the nonparametric density esti-
mation, i.e., qI ðhsys Þ ¼ qI ðMðhÞÞ.
(3) Calculate the importance weights as
J. GUO et al.
 1a
qO ðMðhÞÞ
w1 ¼ LI ðhÞ
qI ðMðhÞÞ
 1a ð10Þ
q ðMðhÞÞ
w2 ¼ O LI ðhÞLO ðMðhÞÞ
ð8Þ qI ðMðhÞÞ
where w1 and w2 are the weights for posteriors using the sub-
system data and both the subsystem and system data, respec-
tively, while LI ðhÞ and LO ðMðhÞÞ are the likelihood functions
of the subsystems and system, respectively.
distribution of h and the importance weight w1 or w2 as
follows:
q~I ðhÞ ¼ qI ðhÞ  w1 or q~I ðhÞ ¼ qI ðhÞ  w2 ð11Þ
The SIR algorithm may not work well due to the issue of
centered large importance weights which occurs occasionally.
In other words, a small number of large importance weights
can dominate the re-sampling process. Raftery and Bao38 dis-
cussed the centered sampling issue in sampling HIV population
via SIR. In order to address this issue, an adaptive importance
sampling scheme has been suggested in the literature. Liang39
extended the pruned-enriched Rosenbluth method by imposing
upper bounds on weights to limit the usage of certain samples.
The enrichment method distributes samples with extremely
large weights evenly to samples with smaller weights. By split-
ting large weights into smaller ones, a wider range of samples
is explored, and the importance sampling estimator is still unbi-
ased. In the proposed Bayesian melding method for system reli-
ability assessment using the SIR method, weights are functions
of the induced system prior, native prior, and likelihood func-
tions of both the system and subsystems, as shown in Eq.
(10). One key parameter of the adaptive enriched scheme is to
set the threshold weight, which needs to be pre-determined.
Yuan and Druzdzel40 proposed a cumulative weight method
to initialize threshold weights. More specifically,
(1) Generate weight samples using Eq. (10);
(2) Assign m rejection nodes which are randomly selected
from the samples generated in step 1, and track cumula-
tive weights for every rejection node;
(3) Sort these weights ascendingly and assign a threshold
weight tw ¼ wbcmc , where c is denoted as the threshold
percentage;
(4) Examine all samples generated and split the weight of a
rejection node into k ¼ bw1 =tw þ 1c or k ¼ bw2 =tw þ 1c
nodes with smaller weights. A rejection node is defined
(5) Repeat Steps 1–4 until the variance of weights reaches
ð9Þ the pre-determined value. This research uses standard
deviation as the pre-determined threshold value.
4. A case study of Bayesian melding method for an aerospace
system reliability assessment
This section applies the proposed methods for an aerospace
system reliability assessment under different Bayesian informa-
tion integration methods. System reliability assessment out-
comes are compared under four major scenarios based on
the availability of prior information and test data.
A Bayesian approach for integrating multilevel priors and data 47

where hi , ni , si , i ¼ 1; 2; 4a;4b, and 5 are the parameters in Ber-

noulli trials, i.e., the reliability of each subsystem, the number
of total tests, and the number of successes of the ith subsystem,
respectively.
Prior distributions of subsystems can be obtained through
Fig. 4 An aerospace system reliability block diagram (RBD).41 the prior elicitation method discussed in Section 2.1. In this
research, one expert was interviewed to provide prior informa-
4.1. Background tion about the reliabilities of subsystems in the form of mode
estimation and confidence levels (Table 2).
The case study addresses a critical piece of an aerospace system When conjugated priors are used, the posterior distribution
whose reliability block diagram is shown in Fig. 4.41 Subsys- can be obtained analytically. Fig. 5 illustrates the subsystem
tems 1, 2, and 5 are in a series structure, while subsystems 4a prior and posterior reliability distributions simulated through
and 4b are in a parallel structure. It is very costly to implement the MC method, and the density is dimensionless. The system
testing for reliability demonstration, especially at the system prior and posterior reliability, in the sixth panel of Fig. 5, is
level. Only subsystem-level test results and limited system- derived from the subsystem priors and posteriors based on
level historical testing data in terms of either pass or fail are the system configuration. The posterior modal reliabilities of
available. The objective is to estimate the parameter, hsys , of subsystems 1, 4a, and 4b are less than their prior modal relia-
the system reliability by utilizing limited binary testing data bilities due to the observed failures from the subsystem testing
and expert prior knowledge. Table 1 shows subsystems test data (Table 1). The posterior modal reliabilities of subsystems
results, where ‘‘P”, ‘‘F”, and ‘‘NT” stand for pass, fail, and 2 and 5 are very close to their prior modal reliabilities since
not-tested, respectively. The historical statistical data of a there is no failure observed from these two subsystems.
system-level test is 12 passes out of 13 tests. In certain scenarios, system-level prior and data are avail-
able, while subsystem-level prior and data are limited, such
4.2. System and subsystem reliability assessment using single- as the reliability performance of a hydraulic valve of an aero-
level priors and data only space system in a harsh vacuum environment. In such cases,
system reliability can be simply estimated as a single item. In
this example, the system reliability posterior distribution is
Due to the lack of system-level prior information and experi-
betaða þ y; b þ n  yÞ with the beta prior betaða; bÞ, where y
mental data for some aerospace systems, system reliability
and n are the number of passes and the total number of trials,
may have to be assessed completely relying on subsystem-
respectively. A betaða; bÞ is referred as the prior distribution of
level information. The assessment method based on a system
system reliability, which represents prior information centered
structure with subsystem-level prior and data that has been
around hsys , i.e., a 50%–50% fail versus pass probability.
discussed in Section 3.1 can be employed. Eq. (12) below pre-
sents the system reliability and the likelihood function for the
given example and testing data in Table 1.
8 Table 2 Expert estimates on the reliability of each subsystem.
> hsys ¼ h1 h2 h5 ðh4a þ h4b  h4a h4b Þ
>
>      
>
> n1 s1 n1 s1 n2 n5 s5 Subsystem Mode Confidence significance
>
> ¼ h ð1  h Þ h s2
ð1  h Þn2 s2
h ð1  h5 Þn5 s5
>
> L sys 1 2
>
< s1
1
s 2
2
s5 5 1 0.999 10
   
n4a s4a n4b s4b 2 0.952 21
>
>  h4a ð1  h4a Þn4a s4a þ h4b ð1  h4b Þn4b s4b
>
> s s 4a 0.930 10
>
>  
4a
 
4b

>
> 4b 0.990 10
>
> n4a s4a n4b s4b
:  h4a ð1  h4a Þn4a s4a h4b ð1  h4b Þn4b s4b 5 0.950 20
s4a s4b
ð12Þ

Table 1 Test results of studied aerospace system.

No. Subsystem 1 Subsystem 2 Subsystem 4a Subsystem 4b Subsystem 5
1 P P NT NT NT
2 P P NT NT NT
3 P P NT NT NT
4 F P NT NT NT
5 P P NT NT NT
6 P NT P P P
7 P NT P F P
8 P NT P P P
9 P NT F F P
10 P NT P P P
11 P P P P P
12 P P P P P
13 P P P P P
Sum of passes 12 8 7 6 8
48
Fig. 5 Subsystems and system posterior reliabilities using the subsystem priors and data only.
Using the conjugate beta distribution, the posterior distribu-
tion of hsys follows an updated beta distribution, i.e.,
pðhsys jYsys Þ  betað13; 4Þ, where Ysys denotes the system-level
test data.
4.3. System reliability estimation using priors and data from
system % and/or subsystems
In this section, the focus is to integrate system-level reliability
information with subsystem-level reliability information. In
particular, updating the posterior system reliability by pooling
the system prior and subsystem priors. The subsystem poste-
rior reliability is also updated by integrating the system-level
reliability information through the Bayesian melding method.
4.3.1. System and subsystem reliability estimation using
subsystem prior and data as well as system prior
In cases of the existence of subsystem prior and test data as
well as the system prior reliability, the subsystem posterior dis-
tribution can be evaluated based on the updated subsystem
prior and the subsystem-level data. Firstly, the updated sub-
system prior distribution is evaluated as in Eq. (8). Then, an
J. GUO et al.
updated subsystem posterior distribution can be estimated
using the updated subsystem prior distribution and the
subsystem-level test data. The system posterior distribution
can be further derived from the updated subsystem posteriors
based on the system structure. Considering the computational
advantage of the SIR algorithm over the Metropolis-Hastings
method and the separable sampling weights in the posterior
inference as in Eq. (10), the adaptive SIR simulation algorithm
discussed in Section 3.2.2 is applied to make inference on the
subsystem posterior reliability. The number of original sam-
pled importance weights is set to 105, and the number of
selected weights and the percentage parameter for determining
the threshold weight value are assigned as 1000 and 90%,
respectively. The weight splitting process stops when the stan-
dard deviation of importance weights is smaller than 100.
Fig. 6 shows the subsystem posterior from the updated prior
and system posterior derived from subsystem posteriors based
on the system configuration. For subsystem reliability infer-
ence, when system prior information is not incorporated for
updating the subsystem prior, the subsystem posterior reliabil-
ity may be overestimated. Fig. 7 shows the updated system
prior, the system posterior derived by using the system-level
prior information, and the system posterior reliability without
A Bayesian approach for integrating multilevel priors and data
Fig. 6 System and subsystem reliability posteriors based on system prior and subsystem prior and data.
Fig. 7 System posteriors reliability using various-level prior and
data.
using the system-level prior information in Section 4.2. The
system posterior modal reliability of 0.521 by integrating the
system-level prior with the subsystem prior and data is lower
than the system posterior modal reliability of 0.783. This is
found using the subsystem prior and data due to the neutral
system prior specification of betað2; 2Þ.
4.3.2. System and subsystem reliability estimation using system
prior and data as well as subsystem priors
When the system prior reliability, system-level test data, and
the subsystem prior reliability are available, the posterior dis-
49
Fig. 8 Native, induced, pooled, and updated system prior.
tribution of system reliability can be evaluated by integrating
the pooled system prior reliability with the likelihood function
for the system-level data. The pooled system prior distribution
is the weighted average of the system native prior and induced
system prior reliability as in Eq. (6). Thus, the system posterior
reliability based on the pooled system prior integrates the sys-
tem prior, system data, and subsystem priors. The simulation
method for no-closed form system posterior reliability infer-
ence uses the MCMC method by introducing proposal distri-
butions. Since it is difficult to obtain full conditionals using
50
Fig. 9 System posterior reliability using subsystem prior and
data, and using the pooled and updated system priors and system
data.
Eq. (12), the MH algorithm is suggested with the pooled sys-
tem prior reliability as the proposal distribution. Fig. 8 com-
pares the system native prior distribution, the induced
system prior distribution, and the pooled system prior distribu-
tion as well as the updated system prior distribution derived
from the updated subsystem priors through the system config-
uration. It is noted that the induced system prior represents the
Fig. 10 System and subsystem priors and posterior reliabilities using both system- and subsystem-level priors and data.
J. GUO et al.
prior information from subsystems only and does not include
the native system prior information. The pooled system prior
distribution includes both the native and induced system prior
information, while the updated system prior integrates the
updated subsystem priors through the system structure func-
tion. It is observed (Fig. 8) that both the pooled and updated
system priors balance the native and induced system prior
information. Fig. 9 compares the system reliability posteriors
for the case of using subsystem priors and data only, which
uses the pooled and updated system priors with system data.
According to the derivation of the pooled and updated system
prior distributions (Section 3.2), the lower posterior reliability
estimate through the updated system prior compared to that
using the pooled system prior indicates that the system prior
knowledge has an impact on the subsystem prior update and
the system-level posterior reliability inference.
4.3.3. System and subsystem reliability assessment based on
priors and data from both system and subsystems
In cases when priors and test data at both system and subsys-
tem levels are available, the Bayesian melding method dis-
cussed in Section 3.2 is an effective way to incorporate
multilevel information. Eq. (5) is considered as the determinis-
tic function MðÞ with inputs being the reliabilities of subsys-
tems, h ¼ ½h1 ; h2 ; h4a ; h4b ; h5 T . The pooled prior is obtained in
Eq. (6) by assigning a pooling weight a as 0.5. The posterior
A Bayesian approach for integrating multilevel priors and data
Fig. 11 System and subsystem posterior reliability comparison between the four studied scenarios.
inference needs to evaluate multiple quantities such as the
induced prior, the pooled prior, and the complex likelihood
function. In this case, considering the computational efficiency
of sampling subsystem priors and evaluating system priors and
likelihood functions, the same adaptive SIR algorithm used in
Section 4.3.1 is used to make inferences on system and subsys-
tem posterior reliabilities. Fig. 10 shows the updated prior and
posterior distributions of subsystem and system reliabilities
with both system- and subsystem-level test data. Subsystem
and system posterior modal reliabilities are less than those
using subsystem priors and data as well as the system prior,
which can be attributed to the integration of system-level test-
ing data, i.e., 12 successes out of total 13 tests.
For comparison purpose, the posterior system and subsys-
tems reliability distributions evaluated using various available
multilevel priors and data are plotted in Fig. 11. For the
subsystem-level posterior reliability assessment (the first 5 pan-
els in Fig. 11), the modal reliability with only subsystem prior
and data is the highest due to higher subsystem prior specifica-
tions. When the system prior information is integrated in addi-
tion to the subsystem prior and data, the modal reliability
shifts to the left due to a neutral system prior specification.
In addition, when system-level data are also available, for sub-
systems 1, 4a, and 4b, due to the observed failed tests, the
modal reliability shifts to the furthest left. Since there are no
observed failures for subsystems 2 and 5, the modal reliabilities
51
of these two subsystems are very close when using the system
prior only in the case of using system prior and data.
At the system level (the 6th panel in Fig. 11), only system-
level data are available. The two system posterior reliability
distributions under both pooled and updated system priors
are located to the left of the four posterior distributions. This
indicates a relative low system success rate, i.e., 12 successes
out of a total of 13 tests, can lead to lower posterior system
reliability assessment. When only subsystem-level data are
available, the posterior system reliability distribution shifts to
the right, which can be explained as a higher prior reliability
specification at the subsystem level. In summary, the BM
method provides a rational mechanism to integrate multiple-
level system and subsystem priors and data for system and sub-
system posterior reliability inferences.
5. Conclusions
This paper investigates the applications of various Bayesian
methods for system reliability assessment when multilevel reli-
ability information is available. Particularly, four main scenar-
ios in terms of the availability of system and/or subsystem
priors and data are discussed. Specific Bayesian inference
methods are recommended for updating both system- and
subsystem-level posterior reliability distributions. The Baye-
52
sian melding method is proposed to update the subsystem pos-
terior reliability using the proposed SIR simulation algorithm,
which has less computational complexity compared to that of
the MCMC algorithm when complex priors and likelihood
functions are involved. System reliability assessment outcomes
are compared using different levels of prior and data informa-
tion available through the MCMC and adaptive SIR simula-
tion algorithms. Through this study, it is observed that
different Bayesian information fusion mechanisms for
system- and subsystem-level priors and data can lead to differ-
ent posterior reliability estimations. When multiple levels of
information are integrated through the Bayesian melding
method, the reliability inferences are more accurate. However,
misspecified information can also overestimate or underesti-
mate reliabilities at both system and subsystem levels.
In the literature, updating subsystem posterior reliability by
using system-level prior and data has not been investigated.
The studied Bayesian methods using varying multilevel relia-
bility information can be useful for more accurate and efficient
system and subsystem reliability evaluation and prediction.
The proposed SIR simulation algorithm is applicable for
larger-scale system reliability assessment when complex priors
and likelihood functions exist. The current work focuses on
commonly used binary data, and future work will be extended
to multiple types of test data such as lifetime data and degra-
dation data.
References
1. Sander P, Badoux R. Bayesian methods in reliability. Dor-
drecht: Springer; 1991.
2. Hamada M, Wilson A, Reese CS, Martz H. Bayesian reliabil-
ity. New York: Springer Science+Business Media; 2008.
3. Springer M, Thompson W. Bayesian confidence limits for the
product of N binomial parameters. Biometrika 1966;53(3/4):611–3.
4. Springer M, Thompson W. Bayesian confidence limits for system
reliabilityProceedings of the 1969 annual reliability and maintain-
ability symposium. New York: Institute of Electrical and Elec-
tronics Engineers; 1969. p. 515–23.
5. Tang J, Tang K, Moskowitz H. Exact Bayesian estimation of
system reliability from component test data. Nal Res Log 1997;44
(1):127–46.
6. Singh H, Cortellessa V, Cukic B, Gunel E, Bharadwaj V. A
Bayesian approach to reliability prediction and assessment of
component based systemsProceedings on the 12th international
symposium on software reliability engineering. Piscataway
(NJ): IEEE; 2001. p. 12–21.
7. Mastran D. Incorporating component and system test data into
the same assessment: a Bayesian approach. Oper Res 1976;34
(3):113–20.
8. Mastran D, Singpurwalla N. A Bayesian estimation of reliability
of coherent structure. Oper Res 1978;26(4):633–72.
9. Martz H, Waller R, Fickas E. Bayesian reliability analysis of series
system of binomial subsystem and components. Technometrics
1988;30(2):143–54.
10. Martz H, Waller R. Bayesian reliability analysis of complex
series/parallel systems of binomial subsystems and components.
Technometrics 1990;32(4):407–16.
11. Winkler R. The consensus of subjective probability distribution.
Manage Sci 1968;15(2):B61–75.
12. Hamada M, Martz H, Reese CS, Graves T, Johnson V, Wilson A.
A fully Bayesian approach for combining multilevel failure
information in fault tree quantification and optimal follow-on
resource allocation. Reliab Eng Syst Safe 2004;86(3):297–305.
J. GUO et al.
13. Bier V. On the concept of perfect aggregation in Bayesian
estimation. Reliab Eng Syst Safe 1994;46(3):271–81.
14. Johnson V, Graves T, Hamada M, Reese CS. A Hierarchical
model for estimating the reliability of complex systems. Bayes Stat
2003;7(2):199–214.
15. Wilson A, Graves T, Hamada M, Reese CS. Advances in data
combination, analysis and collection for system reliability assess-
ment. Stat Sci 2006;21(4):514–31.
16. Poole D, Raftery A. Inference for deterministic simulation models:
the Bayesian melding approach. J Am Stat Assoc 2000;95
(452):1244–55.
17. Reese CS, Wilson A, Guo J, Hamada M, Johnson V. A Bayesian
model for integrating multiple sources of lifetime information in
system-reliability assessments. J Qual Techol 2011;43(2):127–41.
18. Guo J, Wilson A. Bayesian methods for estimating system
reliability using heterogeneous multilevel information. Technomet-
rics 2013;54(4):461–72.
19. Rubin D. Bayesianly justifiable and relevant frequency calcula-
tions for the applied statistician. Ann Stat 1984;12(4):1151–72.
20. Evans M, Swartz T. Methods for approximating integrals in
statics with special emphasis on Bayesian integration problems.
Stat Sci 1995;10(3):254–72.
21. Robert C. Bayesian computational methods. In: Gentle J, Härdle
Y, Mori Y, editors. Handbook of computational statistics (volume
I) concepts and fundamentals. Heidelberg: Springer; 2004 [chapter
III.11.11].
22. Geman S, Geman D. Stochastic relaxation, Gibbs distributions
and the Bayesian restoration of images. IEEE T Pattern Anal
1984;6(6):721–41.
23. Metropolis N, Rosenbluth A, Rosenbluth M, Teller A, Teller E.
Equation of state calculations by fast computing machines. J
Chem Phys 1953;21(6):1087–91.
24. Hastings W. Monte Carlo sampling methods using Markov Chain.
Biometrika 1970;57(1):97–109.
25. Liang F, Liu C, Carroll R. Advanced Markov Chain Monte Carlo
methods. Chichester: John Wiley and Sons; 2010.
26. Bedford T, Quigley J, Walls L. Expert elicitation for reliable
system design. Stat Sci 2006;21(4):428–50.
27. Quigley J, Bedford T, Walls L. Prior distribution elicitation,
encyclopedia of statistics in quality and reliability. Hoboken: John
Wiley & Sons; 2008.
28. Weiler H. The use of incomplete beta functions for prior
distribution in binomial sampling. Technometrics 1965;7
(3):335–47.
29. Meyer M, Booker J. Eliciting and analyzing expert judgment: a
practical guide. Philadelphia: ASA-SIAM; 2001.
30. Rosenthal J. Optimal proposal distribution and adaptive MCMC.
Handbook of Markov Chain Monte Carlo. Boca Raton: Chapman
& Hall/CRC; 2011 [chapter 4].
31. Chib S, Greenberg E. Markov Chain Monte Carlo simulation
methods in econometrics. Econ Theor 1996;12(3):409–31.
32. Raftery A, Givens G, Zeh J. Inference from a deterministic
population dynamics model for bowhead whales. J Am Stat Assoc
1995;90(430):402–30.
33. Wolpert R. Comment on ‘‘Inference from a deterministic popu-
lation dynamics model for bowhead whales”. J Am Stat Assoc
1995;90(430):426–7.
34. Genest C, McConway K, Schervish M. Characterization of
externally Bayesian pooling operators. Ann Stat 1986;14
(2):487–501.
35. Taddy M, Lee H, Sanso B. Fast inference for statistical inverse
problems. Inverse probl 2009;25(8):085001.
36. Rubin D. Using SIR algorithm to simulate posterior distributions,
Bayesian statistics. Oxford: Clarendon Press; 1988.
37. Frigessi A, Martinelli F. Computational complexity of Markov
Chain Monte Carlo methods for finite Markov random fields.
Biometrika 1997;84(1):1–18.
A Bayesian approach for integrating multilevel priors and data
38. Raftery A, Bao L. Estimating and projecting trends in HIV/AIDS
generalized epidemics using incremental mixture importance
sampling. Biometrics 2009;66(4):1162–73.
39. Liang F. Dynamically weighted importance sampling in Monte
Carlo computation. J Am Stat Assoc 2002;97(459):807–21.
40. Yuan C, Druzdzel M. Improving importance sampling by adap-
tive split-rejection control in Bayesian network, advances in
53
artificial intelligence. 20th conference of the Canadian society for
computational studies of intelligence, Canadian AI 2007; 2007
May 28–30, Montreal, Canada; 2007. p. 332–43.
41. Li Z, Mense A. Bayesian reliability modeling for pass/fail systems
with sparse data. The 61st annual reliability & maintainability
symposium. Palm Harbor: IEEE Press; 2015. p. 1–7.