Professional Documents
Culture Documents
6. AI-2020-Lee-Analysis of the Reliability
6. AI-2020-Lee-Analysis of the Reliability
A R T I C LE I N FO A B S T R A C T
Keyword: The reliability of a starter-generator in transport aircraft was assessed. Using the process of reliability-centered
Bayesian Network maintenance (RCM), necessary decisions were made not only to satisfy the reliability requirement but also to
Reliability-Centered Maintenance reduce the maintenance load. Failure data have indicated that the life of a starter-generator is limited by the
Starter-Generator reliability of a bearing. The degradation of the bearing was represented by a dynamic Bayesian network (DBN).
Bearing
Parameters were learned by using the EM algorithm given failure data. The DBN model yielded more con-
servative risk projection than traditional survival analysis due to the limited number of failure data. The DBN
model can make up for the lack of data records by knowledge of experts. Using a calibrated model, the time for
inspection was determined to maintain reliability over a prescribed amount of time.
1. Introduction the JSSG 2006 specifies an acceptable level of 1 × 10−7 per flight hour.
Quantitative risk assessment ability is a key factor of RBI im-
Risk and reliability based inspection planning (RBI) is a strategy to plementation. Risk quantification requires that models accurately assess
reduce the total cost (sum of the cost of risk and maintenance), through the risk due to degradation, inspection ability (probability of detection),
optimization of the inspection schedule. The risk cost includes the cost and environmental factors (load, contamination, humidity etc.) [2].
of aircraft downtime (due to failure), damage, or loss of the aircraft. RBI Due to limited knowledge of the input parameters for such models, a
procedures have been developed and implemented for aircraft since the probabilistic consideration is necessary, and advanced statistical
1980s, largely to inspect the fatigue deterioration of structural elements methods such as the Bayesian Network (BN) are commonly used to
[1–3]. Aircraft structures are designed with a damage tolerance ap- model the deterioration data [11]. Degradation mechanisms such as
proach that is suitable for RBI, because the assumption of the initial fatigue crack propagation (usage-dependent) or stress corrosion
crack size is relatively conservative. With methods of conventional in- cracking (time-dependent) can be modeled using the BN [5,6,11]. Much
itial crack size assumption, it is likely that a crack could grow un- of the popularity of the BN can be attributed to the existence of efficient
realistically fast, thus requiring frequent inspection. With field inspec- and robust procedures for learning parameters from observations [12].
tion data, the initial crack size distribution can be calibrated and It is quite common that the data available for calibrating the prob-
validated [4–6]. After calibration, inspections with no-detection gen- abilistic model is incomplete. For example, missing data can occur,
erally yield smaller initial crack size distributions, and the risk of during an aircraft inspection, where the results generally include binary
fracture is lowered. Inspections can be conducted whenever the re- detection data without the intensity of signal and noise. The expecta-
sulting risk is equal to or below the allowable level. tion and maximization (EM) algorithm enables parameter estimation
RBI has brought great flexibility to inspection planning [6,7]. By for probabilistic models with incomplete data [12].
setting the allowable risk, which generally results in an increased cost As with aircraft structures, the service life of a bearing is mostly
of risk, the cost of maintenance is decreased. As a result, the total cost is limited by its fatigue strength. Due to uncertainties in determining the
minimized [8]. The methods used to set safe lives or inspection inter- material constants for life prediction, a probabilistic approach of life
vals are generally based on one or more internationally recognized estimation was applied as early as the 1920s [13]. The L10 life, which is
airworthiness standards such as DEFSTAN 970 [9] or the Joint Services the amount of time that 90 % of a group of bearings will complete or
Specification Guide (JSSG) [10]. DEFSTAN 970 specifies safety limits in exceed (without failing due to rolling-element fatigue), is the basis for
terms of a total probability of failure of 1 × 10−3 for an aircraft, whereas calculating bearing life and reliability.
⁎
Corresponding author.
E-mail address: dongsuchoi@mnd.go.kr (D. Choi).
https://doi.org/10.1016/j.ress.2019.106628
Received 13 February 2019; Received in revised form 12 July 2019; Accepted 31 August 2019
Available online 26 September 2019
0951-8320/ © 2019 Published by Elsevier Ltd.
D. Lee and D. Choi Reliability Engineering and System Safety 195 (2020) 106628
Although the life of a bearing is affected by many factors, it is Pr(a, e , c ) Pr(a)Pr(e | a)Pr(c | a)
Pr(a, c | e ) = =
predicted using only the dynamic load capacity and applied load on the Pr(e ) ∑a Pr(a)Pr(e | a) (2)
bearing. The concept of rolling bearing rating life and basic load rating
(load carrying capacity) were introduced by A. Palmgren in 1937 [14]. As a result, marginal posterior probabilities of A and C are also updated.
Note that without having any information about A, B and C are de-
The 1952 Lundberg and Palmgren report on ‘Dynamic Capacity of
Rolling Bearings’ is still the basis for all bearing life calculations [15]. pendent. Provided that A is known, B and C are independent because
communication between B and C is blocked by A. For a given set of
At the time of Palmgren, the dominant failure mode was subsurface
cracks initiated at voids and inclusions [16]. The basic rated life and the evidence, it is possible to infer the dependence assumptions encoded in
the graphical structure using the rules of d-separation [19].
dynamic load rating represent the subsurface fatigue performance of
the bearing. The improvements in bearing fabrication techniques sig- A dynamic Bayesian network (DBN) is a special class of BN, which
represents stochastic processes. A group of BNs (slices) consists the
nificantly reduced the risk of subsurface cracks. Instead, lubrication and
contamination are now the two most important modes of failure [17]. DBN, each of which represents the state of ith time step. From i to i + 1,
slices are connected by acyclic edges. A DBN is called homogeneous as
In principle, the same dynamic load rating is obtained for bearings
with the same internal geometry, although they may have different far as the model structure and the conditional probability tables are
unchanged. A number of inference algorithms for DBN were developed
surface microgeometry, waviness, raceway, rolling element profilo-
metry, shape, internal precision tolerances, material fatigue strength, [21], and software packages are also available.
and type of heat treatment [16]. Thus, ISO 281:2007 introduced a life
modification factor, aISO, which is a function of the lubrication regime, 2.2. Expectation maximization algorithm
contamination, and fatigue stress limit. Further, aISO is a nonlinear life
modification factor, and may result in an infinitely long bearing service The EM is used to estimate parameters in probabilistic models with
life. The benefit of introducing this factor is that it considers environ- incomplete data. The EM algorithm alternates between the steps of
mental effects such as lubrication and contamination, which are the two guessing a probability distribution over completions of missing data
dominant failure modes of bearings. The use of aISO has been ex- given the current model (known as E-step) and then re-estimating the
tensively debated, especially for the existence of the fatigue load limit model parameters using these completions (known as the M-step) [12].
[13,16,18]. However, evidences of the benefits of the ISO life mod- During E-step, one does not usually need to form the probability dis-
ification factor has been accrued over many years [16]. tribution over completions explicitly, but rather need only to compute
In this study, the RBI was developed for a starter-generator using the ‘expected’ sufficient statistics over these completions. E-step makes the
process of reliability-centered maintenance (RCM), and the modified objective function has one global maximum that can be computed in
rating life model provided in ISO 281:2007. Field inspection records closed form. M-step comes from the fact that model re-estimation can
were used for calibration and validation, and the assumed inspection be thought of as ‘maximization’ of the expected log-likelihood of the
capability was considered for risk assessment. Suitable actions were data.
recommended to solve operational difficulties. More formally, the EM algorithm augments the observed data, D,
with latent data, Z, so that the augmented posterior distribution p(θ | Z,
D) is simple [22]. In the most general setting, the E-step consists of
2. Theoretical Background computing
In the following, a brief introduction to BN is given, limited to the conditional predictive distribution of the latent data which is condi-
case of discrete random variables, i.e., random variables that are de- tional on the current guess of the posterior mode. The Q function is the
fined in a finite space. expectation of log [p(θ | Z, D)] with respect to p(Z | θi, D). In the M-step
The BN is a directed acyclic graph (DAG), which is described by the Q function is maximized with respect to θ to obtain θi + 1.
nodes and edges [19]. Nodes represent random variables and edges As with most optimization methods for nonconcave functions, the
denote dependencies among them [20]. Fig. 1 shows a simple BN, EM algorithm comes with guarantees only of convergence to a local
where node ’A’ is the parent of node ’B’ and ’C’, and both ’B’ and ’C’ are maximum of the objective function (except in degenerate cases).
the children of ’A’, A node with no parent has its own probability dis- Running the procedure using multiple initial starting parameters is
tribution, and a node with parents has a joint probability distribution of often helpful; similarly, initializing parameters in a way that breaks
itself and its parents. The joint probability mass function (PMF) of this symmetry in models is also important [12].
network (Fig. 1) is given by the probability chain rule.
2.3. Rolling bearing life prediction
Pr(a, b, c ) = Pr(a)Pr(b | a)Pr(c | a) (1)
The Lundberg and Palmgren theory [15] developed the basis for the
where Pr(b | a) is the conditional PMF of B given A, and so is Pr(c | a) . calculation of the dynamic load rating and equivalent dynamic load of
Evidence update the BN. For example, when the state B in the rolling bearings as it is applied today in the ISO 281 [23] basic rating
network in Fig. 1 is observed to be e, this information propagates life equation:
through the network and the joint PMF of A and C changes according to
p
Bayes’ rule to C
L10 = ⎛ ⎞
⎝P⎠ (4)
where L10 is the rated fatigue life, at 90 % reliability in million re-
volutions, C is the basic dynamic load rating or capacity, P is the
standardized dynamic equivalent load or applied load, and p is the life
equation exponent. The basic dynamic load rating, C, is defined as a
calculated constant radial load which a group of identical bearings can
theoretically endure for a rating life of one million revolutions. The
standardized dynamic equivalent load, P, is defined as the equivalent
Fig. 1. Simple BN. radial load on the bearing which will result in the same life as the actual
2
D. Lee and D. Choi Reliability Engineering and System Safety 195 (2020) 106628
3
D. Lee and D. Choi Reliability Engineering and System Safety 195 (2020) 106628
Table 2
Starter-generator inspection data with the detection of failure indications.
Year Time Type Year Time Type Year Time Type
examining the maintenance records, which have been collected since inversely proportional to stress, it will generally be lower for a hybrid
2004, 166 cases of starter-generator failure were found (Table 2). bearing than for a full complement steel bearing, as the elastic modulus
Within these 166 cases, 24 starter-generators required replacement, and of ceramic is greater than that of steel in most cases [25]. The dynamic
of these 24 cases, 21 starter-generators were exchanged due to bearing load rating of hybrid ceramic bearings with rolling elements made of
defects. Thus, it is reasonable to conclude that the failed state is largely silicon nitride is approximately 70 % that steel bearings [25].
caused by bearing failure. There are four types of failure consequences: hidden failure con-
There are two bearings in the starter-generator, both of which are sequences, safety and environmental consequences, operational con-
hybrid ceramic bearings with ceramic balls. An equivalent bearing with sequences, and non-operational consequences [24]. The failure of the
steel balls has a basic dynamic load rating of 13.5 kN and a fatigue load starter-generator affects operations. The starter-generator is not a flight
limit of 0.28 kN (www.skf.com). Ceramics have a higher modulus of safety-critical part. The only consequence of its failure is an aborted
elasticity than high carbon steel. Therefore, in the ceramic bearings, the mission.
smaller deformation of the rolling elements (balls or rollers) generates a The aircraft electrical system consists of two starter-generators and
higher stress at the contact point between the rolling elements and the four batteries. If both starter-generators fail, the two main batteries can
raceway, when compared to a steel bearing. Because the life a bearing is provide electrical power for 30 min. Additionally, there are two backup
4
D. Lee and D. Choi Reliability Engineering and System Safety 195 (2020) 106628
5
D. Lee and D. Choi Reliability Engineering and System Safety 195 (2020) 106628
the starter-generator life. Because bearings are the weakest link of the
the viscosity ratio (κ) also need to be considered. Other variables, the
starter-generator, it was assumed that the life of the starter-generator is
contamination factor (eC) and the fatigue load limit (Cu) were con-
proportional to the life of the bearing described in Eq. (6).
sidered to be constant.
LSG ∝ L10 (11) To predict the life of a starter-generator correctly, it was required to
obtain the accurate measurements of both the dynamic load capacity
Sensitivity analysis has been conducted to choose significant vari- and the applied load, which show variation due to factors like mea-
ables of the starter-generator life estimation model. Variance-based surement error and the inhomogeneity in materials. Because few air-
methods using Sobol’s variance decomposition approach and an esti- craft were in operation, it was impossible to measure the capacity and
mated procedure was employed for this analysis [26]. Probable ranges the load experimentally. Prior distributions with best guesses were used
for each variable are listed in Table 3. It was assumed that the dynamic instead, to get the life distribution using the DBN model. Those priors
load capacity of the equivalent steel ball bearing is the maximum, and were updated with the field inspection data. A calibrated model was
the minimum is 70 % of the maximum dynamic load capacity [25]. The used to obtain the life distribution and the reliability of the starter-
ratio of the dynamic load capacity and the applied load is ranged from 2 generator.
[16] to 20. The maximum ratio of 20 yields about 10 times longer basic We represent the starter-generator life as a DBN model (Fig. 8).
rating life than targeted time of operation (1000 h at 12000 RPM). Similar models were used by the authors to predict stress corrosion
The bearing of the starter-generator is lubricated by grease. The crack and fatigue crack length [5,6]. Nodes (circles) in Fig. 8 represent
accepted view of grease behavior is that it acts as a reservoir releasing variables, and each node has its own conditional probability distribu-
oil into the rolled track and thus replenishing the contact [27]. The tion (CPD). In the DBN model, each slice is the snapshot of the system at
main problem of grease lubrication is that the degradation of the film a given time. The subscript N for each node denotes that the node be-
thickness occurs much earlier than the life of a bearing [28]. κ is the longs to Nth slice. CPN node represents the ratio of the dynamic load
ratio of the viscosity of grease base oil and the reference kinematic capacity and the applied load. κN is the node for the viscosity ratio, and
viscosity. In general, the kinematic viscosity of grease base oil is higher eC,N node represents the contamination factor. The life of the starter-
than the reference kinematic viscosity. However, considering the de- generator is denoted by LSG,N node, which contain all the possible states
gradation of the film thickness, it was assumed that κ yields aISO less of it at Nth slice. ZN and RN are nodes for the inspection and the relia-
than the unity given ranges of P, eC, and Cu. Because grease in sealed bility respectively. Among nodes, LSG,N and RN are query nodes. In-
and shielded bearings can provide an effective barrier against dust and spection node, ZN, is the only observable node. The probability dis-
dirt, it was assumed that the contamination level maintains at least tributions of the time-invariant model parameters, CPN, κN, and eC,N, are
normal cleanliness. This condition is typical of bearings greased for life updated with the observations. The posterior distributions of query
and shielded. According to ISO 281:2007, therefore, eC is ranged from variables were obtained from Bayes’ theorem (Eq. (2)).
0.5 (normal cleanliness) to 1 (extreme cleanliness). The fatigue load The inspection node, ZN, has two state: “detected” and “not de-
limit is the material parameter, and in general has the minimum var- tected”. Each inspection process is characterized by its probability to
iance. In this study, 10 % variation was assumed. detect flaws of different sizes so that, given a particular flaw, there is
Fig. 7 shows the result of the sensitivity analysis for the starter- some probability of detection (POD) uniquely defined for that process
generator life, and the applied load (P) has the highest sensitivity in- [29]. Because the life of a component is affected by the occurrence of
dices. In the probabilistic modeling, the dynamic load capacity (C) and flaws, it was assumed that the size of the flaw shows a negative cor-
relation with the life of the component. At a current inspection op-
Table 3 portunity, the starter-generator with shorter life expectancy is more
Probable ranges of variables. likely to have flaws than starter-generator with longer life expectancy.
Name Variable Unit Probable range Given a particular time of operation, RevN, the POD can, therefore,
be expressed as the function of the life of the starter-generator. RevN
Dynamic load capacity C kN 9.45-13.5
was represented explicitly in the DBN model to emphasize that the POD
Capacity load ratio C /P - 2-20
Viscosity ratio κ - 0.1-0.4 curve depends on the time of operation. The POD was expressed using
Contamination factor eC - 0.5-1 the cumulative distribution function (CDF) of the standard normal
Fatigue load limit Cu kN 0.25-0.31 distribution. Thus, two parameters, a mean and a standard deviation,
6
D. Lee and D. Choi Reliability Engineering and System Safety 195 (2020) 106628
⎝ σPOD ⎠ (12)
where Φ is the CDF of the standard normal distribution, LSG is the life of
the starter-generator, tinsp is the time of operation at the inspection,
μPOD(tinsp) is the location parameter of the normal distribution function
at tinsp, and σPOD is the standard deviation of the POD model. In Eq. (12),
the POD increases as the life of the starter-generator decreases, because
the occurrence of flaws is more likely for the starter-generator has a
shorter life. The POD given a particular time of operation at an in-
spection is the same for all inspection process so that, the location
parameter, μPOD, is a linear function of tinsp. In general, two parameters
in Eq. (12) need to be estimated using the data collected from designed
experiments [30]. Because there’s no data available to evaluate the
effectiveness of inspection, the two parameters were calibrated to
match the POF at 600 h of operation from calibrated model to the POF
from the survival analysis (Fig. 6). Fig. 9 shows the POD model for the
analysis, and a possible improvement was also described. It was as- Fig. 9. Inspection effectiveness used for the analysis.
sumed that the mean is equal to tinsp, and the σPOD is 100 for the con-
ventional inspection. starter-generator is the area of the life distribution up to the current
The reliability node, RN, is the binary variable (0: failure; 1: no- time of operation.
failure). As shown in Eq. (9), POF is the complement of reliability. In
tc
node RN, the POF and the reliability are calculated consecutively. The POF = ∫0 LSG (t ) dt (14)
event of failure was determined by the limit state function.
g = LSG (t ) − tc (13) The prior distributions for each node are summarized in Table 4.
Posterior starter-generator life distributions were obtained using the
where tc is the current time of operation. The limit state is defined as probabilistic model depicted in Fig. 8. A “forward-backward” algorithm
g = 0 . This is the boundary between the failure domain and the no- was utilized to conduct exact inferences. Murphy’s Bayesian Network
failure domain in the starter-generator life space. The POF of the Toolbox [21] was used as an inference engine. Inference problems in
7
D. Lee and D. Choi Reliability Engineering and System Safety 195 (2020) 106628
Table 4
Discretization scheme where nstate is the number of states.
Variable Distribution nstate Interval boundaries
8
D. Lee and D. Choi Reliability Engineering and System Safety 195 (2020) 106628
Fig. 12. Calibrated (a) CP0 and (b) κ0 for starter-generator life prediction.
The denominator of Eq. (17) was obtained using Eqs. (18) and (19).
We have, however, no prior knowledge of the distribution of crack
length under H1. Thus, a uniform distribution for f(LSG; N) was used as a
Fig. 13. Comparison of reliability predictions by the Kaplan-Meier (K-M) non-informative prior for H1.
model, the Weibull model, and the DBN model (calibrated and uncalibrated). Assuming that the inspections were conducted in a timely manner
and the result of each inspection is Di, then maintenance data D consists
of D1, D2, D3, ..., Dn. Provided that each inspection was conducted in-
because the effect of priors was still influential in the calibrated model.
dependently, an event observing inspection result (D) is the intersection
In other words, the number of data for the calibration was not enough
of data (Di). The Bayes factor for each inspection data (Di) is Bi. The
to overcome the effect of priors. Because both the Kaplan–Meier model
total Bayes factor can then be calculated using the independence of the
and the Weibull model are data-driven models, the quality of prediction
inspections as follows:
is largely affected by the number of data available. When there are not
enough data records to learn from, data-driven models’ estimates pro- Pr(D | H0 )
B =
vide poor quality of survival prediction [20]. However, the DBN model Pr(D | H1)
can make up for the lack of data records by knowledge of experts. From Pr(D1 ∩ D2…∩Dn | H0 )
engineering perspective, the conservative prediction by the DBN model =
Pr(D1 ∩ D2…∩Dn | H1)
is more appropriate. Pr(D1 | H0 )Pr(D2 | H0 )…Pr(Dn | H0 )
Validation tests the confidence that can be placed on the calibrated =
Pr(D1 | H1)Pr(D2 | H1)…Pr(Dn | H1)
model. Following Sankararaman et al. [4] the calibrated model was
= B1 B2…Bn (20)
validated using a Bayes factor. The Bayes factor is the ratio of the
likelihoods of the two hypotheses for the correctness of the model: H0 The total Bayes factors calculated by Eq. (20) using 8-fold cross-vali-
that the model is correct, and H1 that the model is incorrect: dation are shown in Fig. 14 The confidence of the calibrated model can
be tested from the posterior probability of the null hypothesis, which
Pr(D | H0)
B= determines if “the model is correct”, and is as follows:
Pr(D | H1) (17)
Pr(D | H0)Pr(H0) B
If B is greater than unity, the null hypothesis (H0) can be accepted, and Pr(H0 | D) = =
Pr(D | H0)Pr(H0) + Pr(D | H1)Pr(H1) B+1 (21)
the model has been validated. The numerator of Eq. (17) is proportional
to the integration of the not normalized posterior distribution of crack The higher the posterior probability, the more likely it is that the ca-
length given measurement data. For example, with a perfect prediction, librated model is correct. For example, if the total Bayes factor is equal
the starter-generator life distribution (prior distribution) at rotation N is to the unity, the posterior probability of the calibrated model is 50 %,
the same as the updated distribution (posterior distribution) obtained which indicates that both hypotheses are equally likely. In this context,
by inspection data. In this case, the unnormalized posterior distribution confidence is not the same as confidence intervals in statistical
9
D. Lee and D. Choi Reliability Engineering and System Safety 195 (2020) 106628
10
D. Lee and D. Choi Reliability Engineering and System Safety 195 (2020) 106628
https://doi.org/10.1016/j.engfracmech.2011.02.017. http://linkinghub.elsevier. [17] Glantz W. Contamination in lubrication systems for bearings in industrial gear-
com/retrieve/pii/S0013794411000865. boxes. Ball Bear J 1993(242):20–6.
[5] Lee D, Huang Y, Achenbach JD. Probabilistic analysis of stress corrosion crack [18] TLT. ISO 281:2007 bearing-life standard-and the answer is? Tribol Lubr Technol
growth and related structural reliability considerations. J Appl Mech 2010;66(7):34–36,38–43.
2016;83(2):021003. https://doi.org/10.1115/1.4031899. http:// [19] Jensen FV, Nielsen TD. Bayesian networks and decision graphs. Springer Science
appliedmechanics.asmedigitalcollection.asme.org/article.aspx?doi=10.1115/1. +Business Media, LLC; 2007.
4031899.. [20] Kraisangka J, Druzdzel MJ. Discrete Bayesian network interpretation of the Cox’s
[6] Lee D, Achenbach JD. Analysis of the reliability of a jet engine compressor rotor proportional hazards model. PGM 2014: Probabilistic Graphical Models. 11.
blade containing a fatigue crack. J Appl Mech 2016;83(4):041004. https://doi.org/ 20149783319114330. p. 238–53. https://doi.org/10.1007/978-3-319-11433-0_16.
10.1115/1.4032376. http://appliedmechanics.asmedigitalcollection.asme.org/ http://link.springer.com/10.1007/978-3-319-11433-0_16.
article.aspx?doi=10.1115/1.4032376. [21] Murphy KP. Dynamic Bayesian networks: representation, inference and learning.
[7] Torng TY, Edwards RM, Morgan J. B-1 maintenance schedule impact based on risk University of California, Berkeley; 2002. Ph.D. thesis. http://www.ee.uwa.edu.au/
assessment results. Proceedings of the Aircraft Structural Integrity Program ~roberto/research/projectsbiblio/10.1.1.93.778.pdf.
Conference. 2007. p. 1–28. [22] Tanner MA. Tools for statistical inference. New York, NY: Springer-Verlag New
[8] Straub D, Goyet J, Sorensen JD, Faber MH. Benefits of risk based inspection plan- York, Inc.1461240247; 1993. https://books.google.com/books?id=
ning for offshore structures. Proceedings of 25th International Conference on CAPrBwAAQBAJ&pgis=1.
Offshore Mechanics and Arctic Engineering, Hamburg. 20060-7918-4748-9. p. [23] International Organization for Standardization. ISO 281:2007 rolling bear-
59–68. https://doi.org/10.1115/OMAE2006-92089. http://proceedings. ings–dynamic load ratings and rating life. 2010.
asmedigitalcollection.asme.org/proceeding.aspx?articleid=1598405. [24] Moubray J. Reliability centered-maintenance. New York, NY: Industrial Press Inc.;
[9] U.K. Ministry of Defense. DEFSTAN 00-970 Design and Airworthiness Requirements 1997.
for Service Aircraft. 1983. [25] Zaretsky EV, Chiu YP, Tallian TE. Ceramic bearings for use in gas turbine engines. J
[10] U.S. Department of Defense. JSSG-2006 Aircraft Structure. 1998. Mater Eng Perform 2013;22(10):2830–42. https://doi.org/10.1007/s11665-013-
[11] Straub D. Stochastic Modeling of Deterioration Processes through Dynamic 0726-5.
Bayesian Networks. J Eng Mech 2009;135(10):1089–99. https://doi.org/10.1061/ [26] Saltelli A, Ratto M, Andres T, Campolongo F, Cariboni J, Gatelli D, et al. Global
(ASCE)EM.1943-7889.0000024. http://ascelibrary.org/doi/abs/10.1061/(ASCE) sensitivity analysis: the primer. John Wiley & Sons; 2008.
EM.1943-7889.0000024. [27] Cann PM. Starved grease lubrication of rolling contacts. Tribol Trans
[12] Do CB, Batzoglou S. What is the expectation maximization algorithm? Nature 1999;42(4):867–73. https://doi.org/10.1080/10402009908982294.
Biotechnol 2008;26(8):897–9. https://doi.org/10.1038/nbt1406arXiv:1011. [28] Svoboda P, Kostal D, Kunak J, Krupka I. Study of grease behaviour in a starved
1669v3. https://www.nature.com/nbt/articles?type=primer. lubricated contact. MM Sci J 2014:465–9.
[13] Zaretsky EV. Rolling bearing life prediction, theory, and application. Tech. Rep., [29] Committee on Nondestructive Testing of Aerospace Systems. Economic and man-
NASA/TP2013-215305. NASA Glenn Research Center; 2013. agement aspects of nondestructive testing, evaluation, and inspection in aerospace
[14] Palmgren A. On the carrying capacity and life of ball bearings. Ball Bearing J manufacturing. Tech. Rep.. National Materials Advisory Board; 1977. https://doi.
1937(3):34–44. org/10.1360/zd-2013-43-6-1064.
[15] Lundberg G, Palmgren A. Dynamic capacity of roller bearings. Acta Polytechnica [30] Lee D, Yang S, Park J, Baek S, Kim S. Investigation of detectable crack length in a
Mech Eng Series 1952;2(4). bolt hole using eddy current inspection. Trans Korean Soc Mech Eng A
[16] Gabelli A, Doyer A, Morales-Espejel G. The modified life rating of rolling bearings: a 2017;41(8):729–36.
criterion for gearbox design and reliability optimization. Power Trans Eng [31] Fenton NE, Neil MD. Risk Assessment and Decision Analysis with Bayesian
2015(March):46–54. http://www.powertransmission.com/articles/0315/The_ Networks. Boca Raton, FL: The CRC Press1439809119; 2012.
Modified_Life_Rating_of_Rolling_Bearings:_A_Criterion_for_Gearbox_Design_and_ [32] U.S. Department of Defense. MIL-STD-882E System Safety. 2000.
Reliability_Optimization.
11