Uncertainty reduction using multivariate reliability

models
Muhammad Zubair Qazi Muhammad Nouman Amjad
Department of Nuclear Engineering, Department of Basic Sciences,
University of Sharjah, 27272 University of Engineering & Technology,
Sharjah, UAE. Taxila, Pakistan.
Email: [email protected] Email: [email protected]

Abstract- There are various modeling techniques which are used in factors in a more reliable and belter manner in risk assessment

almost every field of science and engineering to estimate different strategies.

entities on a set of parameters. These estimations contain uncertainty Keywords- Uncertainty, sensitivity, multivariate models,
and imprecision due to non-ideal behavior of input data used in the reliability

modeling techniques. In risk assessment, reliability models are used I. INTRODUCTION

in Probabilistic Safety Assessment (PSA). These models are usually The world has been witnessing the rapid progress in the field
based on some assumptions. They have uncertainties due to of safety, reliability and Digital Instrumentation and Control
inadequate assumptions which must be evaluated accurately to
Systems (DI&C) since the time of Chernobyl disaster.
identi/Y the unambiguous reliability of the safety systems. The
Probabilistic Safety Assessment (PSA) helps to control risk
uncertainty whether it is of model's assumption or model's
and safety factors quite effectively in nuclear industry [ 1 ].
parameter is propagated in the whole risk model. So, the estimation
There are various reliability models which are used in PSA to
of uncertainty is an essential part of measurement in the calculations
of every real world problem particularly in risk assessment. analyze the safety of Nuclear Power Plants (NPPs). These

The two most commonly used techniques which are widely adopted in models are either probabilistic or deterministic depending
different fields for uncertainty evaluation are Monte Carlo upon the nature of the analysis. Probabilistic approach (rather
Simulation techniques and analytical linear approximation than the other approaches) is discussed in this research work.
techniques. The former technique lacks an analytical expression to The motivation to adopt this approach lies in the following
relate input uncertainties to output uncertainties whereas the latter
facts: useful for uncertainty evaluation, common and usual in
technique is limited to linear systems or more precisely the lalter
PSA levels and also useful in Bayes theorem decision making.
technique is limited to weakly nonlinear systems. In order to
The decision theoretic foundations of other approaches are not
overcome this problem, the present research proposes that
yet developed as consistently as the Bayesian decision theory.
uncertainty can be analyzed by using multi-variate reliability models
more compactly and easily which will also provide the degree of Each model is characterized by some of its parameters. The

covariance between estimated parameters. The Uncertainty Analysis parameters are the input data in risk evaluation and can be
Methodology is equally integral i.e. this analysis directly links with estimated from historical failure statistics [2]. Model
the stages needed for parameter estimation. It is used in the ca1culations play a vital role in the present analysis because all
presented work as a sharp tool to reduce uncertainty in model's the decisions related to nuclear safety are made on the basis of
assumption as weil as in model's parameter. In this methodology,
these ca1culations. Usually, these ca1culations have some
Probabilistic approach is used to reduce model's parameter
uncertainty due to imperfect model's assumptions. One has to
uncertainty and sensitivity assessment is used to reduce model's
trust these calculations because there is no experimental data
assumption uncertainty. The proposed research highlights the very
available for some types of failures such as the rare events
important aspect of uncertainty reduction which will help to improve
the effectiveness of PSA by controlling and monitoring of the safety data is scarcely available.

978-1-4673-9073-6/17/$31.00 2017 IEEE

Proceedings of 2017 14th International Bhurban Conference on Applied Sciences & Technology (IBCAST) 361
Islamabad, Pakistan, 10th -14th January, 2017
Fig. 1 shows the comparison between theoretical and practical parameters works on the same principle. This principle uses
approaches for uncertainty evaluation. Theoretically, it is two steps: 1 sI step includes the splitting of total failure
assumed that failures occur in statistically independent manner probability into two parts i.e. dependent failure part and
but practically, some factors like environment related and independent failure part; 2nd step includes the determination of
system related are involved which contradicts the theoretical the relationship between these two parts on a set of parameters
assumption. [4].
B. Origin oj Uncertainty

Theoretically Practically The origin of uncertainty lies in the model

predictions/assumptions on the basis of which the decisions

1 are made. It is commonly said that these uncertainties are of

Reliability Systems two types: i) Parameter uncertainty ii) Model uncertainty. The
Environmental
Models Particular
Factors uncertainty associated with the actual results of the model
Assumption Factors
parameter is known as parameter uncertainty. Model
uncertainty is the uncertainty in the model assumptions and
Failures occur in structure. The former one is usually taken into account
statistically Failures occur in
independent frequently by using various methods/techniques in the
statistically
manner dependent assessment of risk. Whereas, the later one is not usually taken
manner
into account due to the facts that there is: no valid method to

Fig.l. Comparison between Theoretical and Practical approach

represent this type of uncertainty, no clear distinction between
these two different types of uncertainties, no standard
A. Parametrie Models jamiliarization
available due to lack of data.
The multiple effects of different safety systems used in
Studies about dependence aspects of multivariate reliability
industrial area can be estimated by using different models.
models for non-repairable systems have been taken into
Models can be considered as the indicator of the performance
account quite vigorously [5]. Previously, many other authors
of any system. The input data provided to models will be
also studied about implementation of multivariate techniques
inaccurate in spite of the assumptions which are too uncertain
for dependent failures in non-repairable systems. Mostly,
to be true. The output data provided by the model is also
reliability related researches assumes that failures occur in
uncertain due to the uncertain input data. Our models are
statistically independent manner. This type of study does not
always simplifications of the real systems we study.
include dependent failures i.e. common cause failures.
Furthermore, we simply cannot predict precisely about the
Common cause failures make the redundant systems to
future, so we know the model outputs of future conditions are
simultaneous failure. It may be caused due to human errors,
at best uncertain [3]. Shock and non-shock models are two
extreme weather etc. Redundancy is implied in order to
major c1assifications of parametric models. Shock models do
enhance the reliability. Therefore, it is not appropriate to
consider failure mechanism in order to estimate basic event
ignore this kind of dependent failure because of its property of
probabilities whereas, non-shock models do not consider
making simultaneous failure to redundant system [6]. There is
failure mechanism for evaluating basic event probability. The
an inevitable need of an approach which might be able to
objective of all these parametric models is to calculate the
address such correlated failures in reliability study. The
probability of the occurrence of Basic Events on a set of
reliability study without incorporation of this approach (for
parameters. Each parametric model instead of having specific

Proceedings of 2017 14th International Bhurban Conference on Applied Sciences & Technology (IBCAST) 362
Islamabad, Pakistan, 10th -14th January, 2017
dependent failures) will underestimate the failure probability plant specific data. It is evident that safety and risk
thus overestimating the system safety. This uncertainty in management can be improved with generic data [9].
failure probability and system safety is due to the fact that the 11. METHODOLOGY
researchers assume that failures occur in statistically PSA has many levels. All levels essentially include the portion
independent manner. Therefore, dependency in failures must of uncertainty analysis. Fig.2 shows that reliability models are
be taken into account in reliability models in order to have used in all PSA levels. There are two different approaches to
more precise estimates of failure probability and system overcome two different types of uncertainties. PSA ignores
safety. the qualitative analysis in uncertainty evaluation. It uses the
C. Uncertainty Evaluation Techniques probabilistic methodology to predict the uncertainty of safety
There are several approaches which have been used in the past systems quantitatively. The best way to overcome model
such as Correlated life distributions, which apply correlated uncertainty is to perform sensitivity assessment whereas
multi-variate distributions, the Markov model, implicit quantitative analysis followed by the application of probability
method, Multivariate Bernoulli distributions [7] and Taylor distributions are used to overcome model's parameter
series approximation have been developed. The fust three uncertainty.
approaches introduce 2" parameters into a model for a system Bayesian Network includes the prior, likelihood, and posterior
of n components due to which these approaches can be distribution functions. In Fig.2, the overall methodology
referred as CCF modeling techniques. The introduction of 2" depends upon PSA, and Bayesian Network is an essential tool
parameters by these approaches enhances the computational which is frequently used in PSA. Quantitative analysis
complexity thus making these models more complex. The next includes the evaluation of unavailability or basic event
approach mentioned above i.e. multivariate Bernoulli probability followed by the reduction of problem size.
distributions can utilize n2 correlations between each pair of Probabilistic approaches quantify the failure and success
components to explicitly consider the impact of correlated probabilities by the application of probability distributions.
component failures [8]. So these models are less complex than Therefore, Probabilistic approach feed to quantitative analysis.
CCF models because of the introduction of at most a Some important components of pressurized water reactor has
polynomial number of parameters. tested by using multi-variate data testing in Bayesian
Data can be classified as Generic data and specific data. These Network. The following methodology helped to calculate the
data are required to update safety related parameters. Generic uncertainties for multi-variate data testing with the help of
data are easy to access e.g. this type of data can be accessed probabilistic approach. The quantification of probabilities of
through industries. On the other hand, specific data are not important components of PWR through probabilistic approach
easy to access e.g. this type of data can be accessed through is elaborated in section 3.
nuclear plant reliability data system. Uncertainty analysis
Reliability-based design (R BD) is known to provide a rational
(main focus of this paper) in reliability related studies are
basis for incorporating uncertainties in the design environment
usually discredited or doubted because of the scarcity of plant
explicitly into geotechnical design. However, one non
specific data which has to be used in probabilistic safety
terminating criticism of R BD is that there is no particular
assessment for risk management. Plant specific data is
motivation to use it because it seems to produce designs
defmitely superior to generic data. But having said that, it
comparable to existing practice (J Ching).
must not be predicted that safety cannot be improved without

Proceedings of 2017 14th International Bhurban Conference on Applied Sciences & Technology (IBCAST) 363
Islamabad, Pakistan, 10th -14th January, 2017
 PSA
Levels
1
1
1
...

Reliability
Models

1
1

..
U ncertainty
1
- - - - - - .1 _______
Analysis
1 1
1 1
A
... .. 1

Model's Model's
Assumption Parameter

1 Quantitative
1 1
1
1 Analysis

11 Re
: 11 11
edY Re
...

7edY 11 t

1 1
t t
Probabilistic
Probability -------------
Approach
distribution

1
1
1
1 Sensitivity
1--------. - - - ---- - - - ---- - - - ---- - - - ---- - - -
Assessment

Fig.2. Uncertainty Analysis Methodology

Proceedings of 2017 14th International Bhurban Conference on Applied Sciences & Technology (IBCAST) 364
Islamabad, Pakistan, 10th -14th January, 2017
 A. Bayesian Network (distribution estimation) of parameters are focused in risk
Probabilistic models for correlations are necessary for assessment and management.
Bayesian analysis. It is quite easy to obtain updated Bayesian Probabilistic estimation takes advantage of the fact that the
parameters in case of single test data. However, (contrary to data having larger errors spread out in this approach thus
previous one) it is quite difficult to obtain updated Bayesian increasing the effectiveness of utilization of uncertainties.
output parameters in case of multivariate test data. According However, it should be kept in mind that this data may not be
to Bayes theorem, the relation between prior and posterior appropriate for reliability analysis because this data may not
distribution function is as folIows: be complete or it may contain some inaccuracy [ 1 1 ]. The

f(post) K(likelihood)(f(prior))
= (1) failure data for each component with its failure mode is given

Where in Table 1 . The demand failure rate of different components

can be calculated by the following mathematical expression:
f(post) f(y / x)
= (2)
1
f(prior) f(y)= (3) Ad = -A T
r
(6)
2
likelihood f(x / y)
= (4)
Where Ad is demand failure probability and A is failure rate
r
K constant fex)
= = (5) of component. T is the average time span for the surveillance
Where f (x) = Probability Distribution Function (PDF), y is & test interval of the component.
the variable of interest and x is the observed variable, Averaged failure rates for the control and safety systems of

f(post) f(y / x)
= denoted posterior conditional PDF of Pressurized water reactor are given in table 1. Failure rates are
given in per hour. For control rods, failure rate per hour per
y given x, f(prior) f(y)= denotes prior PDF of y,
rod, for failure to enter are given in table 1 , where nf denotes
likelihood f(x / y)
= denotes the likelihood of event y
number of failures and A/h denotes failure rate per hour and
given x, and K constan t fex)
= = denotes the PDF of x. Qd denotes demand probability.
Since the events are dependent on each other within the Table 1: Averaged Failure Rates & Demand Failure Probability Estimation for

Pressurized Water Reactor Safety Systems [11]

Bayesian network, the derivation of posterior probability from
prior probability is a viable method for reliability assessment
Components nf A/h Qd

of complex system. By the use of Bayesian Network, it is

6 3
Pumps (P) 6 10. 10.
possible to perfonn importance and sensitivity analysis of
components and modules through a backward analysis [ 1 0]. 6 3
Control Rods (CR) 4 10. 10.
B. Quantitative measurements

Some important failure parameters related to components such 5 2

Diesels (D) 9 3*10. 3*10.
as hazard rate, repair rate, and failure rate are used in
reliability models as input data. There is always room present
6 3
for comparatively larger errors in risk assessment which is not Valves (V) 32 10. 10.

the case in reliability assessment. Parameter estimates can be

categorized as: point estimates, interval estimates, and
Instruments (I) 6 3*10' lO.j
distribution estimates. Point estimation of parameters are
focused in reliability analysis whereas probabilistic estimation

Proceedings of 2017 14th International Bhurban Conference on Applied Sciences & Technology (IBCAST) 365
Islamabad, Pakistan, 10th -14th January, 2017
 III. R ESULTS AND DISCUSSIONS Table 3: Probability of Reactor Protection System and I & C system for

available and unavailable states [10]

Different failure modes for different Instrwnentation and
System state Probability P(x)
control systems are given in the table 2. In table 2, mean i.e.
Unavailable 0.000061276

,.1, represents failure rate and its unit is taken as /d (per day) or Available 0.99994

/h (per hour). The confidence interval having upper and lower

bounds of 95 % and 5 % is usually good for reducing the size It is shown in Fig. 3 that the confidence interval of number of
of interval thus reducing uncertainties. failures of important control and safety systems of pressurized
Table 2: Failure data for [ & C Systems [12, 13, and [4] water reactor become shorter when more information is used
ID Failure modes Meau ( ,.1, ) in Bayesian Network i.e. multivariate data testing leads to
X;! Transmitter & Pressure sensor failure 0.00000 [7/h shorter confidence intervals whereas uni-variate data testing
Sensor
have significantly larger confidence intervals. The coefficient
X;2 Pressure Sensor/transmitter fail to 0.0001l7/d
of variation can be reduced to approximately 0.2 if
Transmitter operate (FTOP)

X;3 Analog Input Analog input failure 0.000 IId multivariate test data are used instead of uni-variate test data.
X;4 Digital Input Digital input module failure 0.000288/d This is quite significant because the coefficient of variation
XiS Bi-stable Bi-stable fail to operate 0.0005/d
can be as large as 0.35 to 0.45 without incorporation of
Processor (FTOP)
detailed information. It is evident from the above mentioned
X;6 Coincidence Analog (pressure) processor 0.000157/d

Processor logic module fails

ca1culations that bias is reduced whenever more information is

X;7 Digital Output Channel (pressure/temp) 0.0000394/d incorporated. It is also helpful for the purpose of uncertainty
relay fails
reduction i.e. incorporation of more information results in the
X;8 Shunt circuit Breaker shunt failure 0.000581/d
smaller confidence intervals.
X;9 Under Voltage Breaker under voltage 0.000254/d

circuitry failure

X;! 0 Circuit Breaker Circuit breaker fail to 0.0025/d 12

open/close
No.
X Control Rod CRDM failure 0.000015/d
10
0
f
8
Probabilistic availability testing of Reactor Protection System F
a
I & C system has been performed using Bayes theorem. 6 '!-

Bayesian analysis result for reactor trip states are shown in

u 4
table 2. Table 3 specifies available and unavailable states with r
the help of system success and failure probabilities. If system e
s 2
probability is closer to unity then the system is said to be

available and this system state is termed as success state. And, I I I I I I I I
0 I I I

if the system probability is closer to zero, then the system is I

I
I
said to be unavailable and this system state is termed as failure I I

C C D
state. Table 3 shows failure and success probability of Reactor P D P
R R
Protection System where probability of system success i.e.
probability of system availability is 0.99994 whereas the Fig.3: Comparison between Confidence Intervals of number of failures
with uni-variate data testing (Red marked lines) and Confidence Intervals
probability of system unavailability is very much less. of failures with multi-variate data testing (Yellow marked lines)

Proceedings of 2017 14th International Bhurban Conference on Applied Sciences & Technology (IBCAST) 366
Islamabad, Pakistan, 10th -14th January, 2017
 CONCLUSION [4] Muhammad Zubair, Qazi Muhammad Nouman Amjad, "Calculation and

Updating of Common Cause Failures Unavailability by using Alpha Factor

The study has been carried out to analyze the effectiveness of
Model", Annals ofNuclear Energy, vol. 90, pp. 106-114, 2016.
uncertainty reduction by using multivariate reliability models.
[5] R.E. Barlow, F. Proschan, Statistical Theory of Reliability and Life
It is found that the presented technique to reduce uncertainty is Testing: Probability Models, To begin with, Maryland, 1981.

very significant in order to improve the risk and safety [6] H. Joe, "Multivariate Models and Dependence Concepts", Chapman &

Hall, New York, 1997.

management in control and safety systems. With the help of
[7] L. Fiondella, "Reliability and sensitivity analysis of coherent systems with
Bayesian network, it is shown for the case of Pressurized
negatively correlated component failures", International Journal of
Water Reactor safety systems, that the confidence interval Reliability, Quality and Sajety Engineering, vol. 17, pp. 505-529, 2010.

becomes smaller quite significantly whenever more [8] R. Modarres, "High-dimensional generation of Bernoulli random vectors",

Statistics and Probability Letters, vol. 81, pp. 1136-1142, 2011.

infonnation is employed in the uncertainty analysis which
[9] U. Hauptmanns, "The impact of reliability data on probabilistic safety
means that uncertainty is reduced by incorporation of multi
calculations", Journal ojLoss Prevention in the Process lndustries, vol. 1, pp.
testing scheme. Quantitatively, this uncertainty reduction 38-49, 2008.

ranges upto 30-40% in above mentioned calculations which is [10] Khalil ur Rehman, J. Shin, M. Zubair, G. Heo, "Sensitivity Study on

Availability of I&C Components Using Bayesian Network", vol. 2013, 10

quite impressive figure to control risk and safety factors in
pages, 20l3.
nuclear control & safety as weIl as in industrial control &
[lI] U.S.NRC, NUREG-75/014 (WASH-1400), Reactor safety study: an
safety. Direct failures reduce when redundant components are assessment of accident risk in U.S. commercial nuclear power plants,

built in control and safety systems but it leads toward common Washington, OC, USA, 1975.

[12] U.S NRC, NUREG/CR- 5500, Reliability Study: Westinghouse Reactor

cause failures which affects system reliability quite badly.
Protection System 1984-1995, vol. 2, Washington, OC, USA, 1998.
The input values used in the equations are taken from the
[l3] International Atomic Energy Agency (IAEA), "Generic component
observed multivariate test data and as a result the updated reliability data for research reactor PSA," IAEA TECDOC- 0930, Vienna,

output data are obtained. Updated output parameters help us to Austria, 1997.

[14] U.S NRC, NUREG/CR-6928, Industry Averaged Performance for

speculate about reduction in coefficient of variation. This
Components and Initiating Events at U.S Power Plants, Washington, OC,
insight will enable the researchers to design a more reliable
USA, 2007.
and less uncertain safety systems which will truncate the
surplus cost. The study would be carried on further for better
controlling and monitoring of digital I & C systems.
ACKNOWLEDGMENT
The authors are thankful to University of Sharjah, UAE for
providing funding through seed research project no.
(l602040729-P). They are also appreciate the efforts of
University of Engineering & Technology Taxila, Pakistan for
the collaborative research work.
REFERENCES

[ I] USNRC, NUREG/CR- 4350, Probabilistic risk assessment course

documentation. vol. 4, Washington, D.C. USA, 1985.

[2] w Li. Risk assessment of power systems Models, Methods, and

Applications, John Wiley & Sons; 2014.

[3] W Feiler, An introduction to probability and its application, 3'd ed., John

Wiley & Sons; 1968.

Proceedings of 2017 14th International Bhurban Conference on Applied Sciences & Technology (IBCAST) 367
Islamabad, Pakistan, 10th -14th January, 2017
 Analytical analysis of latent heat thermal energy
storage model for solar thermal power plants
Muhammad Zeeshan Khalid*, Muhammad Zubair**, Majid Ali***, Qazi Nouman Amjad*
*Department of Basic S ciences, University of Engineering and technology, 47050 Taxila, Pakistan
**Department of Nuclear Engineering, University of S harjah, 27272, UAE
***US Pakistan Centre for Advanced S tudies in Energy (USPCAS -E), H-12 NUS T, Pakistan

Abstract- In this paper latent heat thermal energy storage based Thermal energy storage (TES ) units have become a key
system was studied using analytical method for solar thermal
area of research. Consumption and depletion of natural
power plant applications. The working fluid circulating in
circular tube by forced convection charges and discharge phase resources such as fossil fuels energy demand increment has
change material based on energy storage unit.Temperature
increased the gap between energy supply and demand. This
distribution and parametric study was performed in order to
study the effects of convection on the melting/solidification of energy gap can be filled by storing excess energy in a suitable
KN03 and NaN03salts.In order to provide the guidelines for
form. There are two types of energy storage, namely long term
system analysis and design optimization, temerature distribtuion
along the axial and radial direction has been obtained and
and short term. If energy is stored for few hours, it is termed
represented in 3d graphs. as short term energy storage and is used for different domestic

Keywords- Energy Storage, Circular tube, Latent heat, Phase

and industrial applications. Energy storage helps in the
change material, Congugate laminar forced convection reservation of natural resources and depletion of capital cost
which leads to the cost effective system [2]. TES devises are
I. INTRODUCTION
generally categorised as sensible heat storage and latent heat
The current economic and technological advancement has storage (LHS ) devices. Although sensible heat storage is
increased the load of energy sourees, many conventional widely used for energy storage in indutrial application, but
energy sources are quite limited and have much hazardous latent heat thermal energy storage (LHTES ) devices have
effect on the health of living beings and environment of our attracted wide range of attraction at industrial and domestic
earth, besides this emission of Green House Gas is considered applications. LHTES provides large energy storage capacity
the main cause of environmental changes. with smaller temperature tluctuations as compare to sensible
In view of these problems a need of an efficient and clean heat storage [3,4].Previous study has showed PCM has 3-4
energy source through the use of advanced technologies and times more heat storage capacity volume than sensible heat
systems are fostering day by day. For this purpose different storage devices in the temperature increment of 20C [5].
types of solar power plant are now being used to convert solar However, due to the low thermal conductivity, supercooling
energy into electricity production unit. These plants are and stability of thermal properties of PCM causes the
constructed in areas which has high solar heat flux and long difficulties for LHTES devices [6].
days to increase the operating capability of solar plants. But Direct solar radiation is thought to be as one of the propective
these plants have one major drawback which is their reduced and effective energy sources and it can be utilized for energy
or no power generation during cloudy or night hour storage for industrial applications. PCM s is one of the
conditions. To overcome this limitation Thermal Energy effective techniques to store solar energy daytime, during
S torage (TES ) unit has also been introduced in many plants which energy is being absorbed and stored by PCM and
[1]. After the energy crisis at 1973-74.Phase change material retrieved during cold hours or night. Lane [7,8] Dincer and

978-14673-9073-6/17/$31.00 2017 IEEE

Proceedings of 2017 14th International Bhurban Conference on Applied Sciences & Technology (IBCAST) 581
Islamabad, Pakistan, 10th -14th January, 2017
Rosen [9] discussed and studied in detail the latent heat was used to simulate the results. Extensive parametric study
storage for solar energy. PCMs are categorized in two groups was performed to study the effect of different parameters
as low temperature and high temperature PCM s. High (such as natuaral convection, mass flow rate of HTF, pipe
temperature PCM s(> 150)such as salt hydrates are used for length etc) on the melting and solidification of paraffin
solar thermal power plants applications. S harma et al [2] (PCM ). The results showed good estimation of PCM melting
reviewed the tested PCM s in literature for solar energy storage and solidification and found to be in good agreement with
applications. exeprimental and numerical resutls.
There are many numerical and experimental studies on the As discussed above, a voluminous literature exists on the
latent heat thermal energy storage (LHTES ) systems. In study of latent heat thermal energy storage system. However
numerical study, Bellici and conti [10] numerically simuated majority of theoretical study was mainly focused on the
the phase change solar shell and tube model using enthalpy system modeling of low temperature PCM s. S o the aim of this
method. S tandard experimental correlations were used to paper is to present high temperature PCM latent heat storage
simulate the forced convective heat charge and extraction. model for solar thermal power plant applications using the
Zhang and Fagri [11] studied semi-analytically latent heat method as applied in Ref. [15]. In this paper cylinder is filled
thermal energy storage system with moderate prandtl number with high temperature PCM (KN03 and NaN03 salts) and
Heat transfer fluid (HTF). Esen et al [12,13] numerically and HTF is passed parallel to it. Parametric study was also
experimentally investigated the cylindrical phase change performed to study the effect of natural convection on the
storage tank attached to solar powered heat pump system by melting of PCM .
using finite difference method. In Ref. [12] PCM is located 11. S OLAR POWER PLANT DESIGN
inside the cylindrical tube and HTF passes parallel to it. In
S olar power plant with latent heat thermal energy storage
Ref.[13] two different models were studied describing the
system model is shown in Fig. 1. HTF in this system absorb
transient behaviour of phase change system. In the first model
heat from sunlight through solar collector. HTF acts as the
PCM is packed in cylinders and HTF pass parallel to it and in
energy source to transfer heat to TES unit to store excessive
the second model pipes contaning HTF ar e embedded in
energy during night or cold hours. HTF is pumped to TES unit
PCM .A number of numerical tests were performed to study
and boiler by two different pathways during sunny hours.
the various geometrical and thermal parameters effect and
Boiler convert fluid into steam and transferred it to steam
storage time.
turbine which is then used to run generator and produce
Tao et al [14] performed numerical study using enthalpy
electricity. This system has rankin cycle which works by
method for thermal energy storage unit in dish solar thermal
converting steams into liquid by condenser at 8 and then again
power generation system. High temperature PCM (molten
pumped to boiler as shown in Fig. 1. In Fig. 1, the numbering
salt) and three kinds of enhanced tubes were used to study and
represents the following components: 1) S olar Collector; 2)
enhanced PCM storage performance. The outcomes showed
Phase Change M aterial; 3) Heat Transfer Fluid; 4) Boiler; 5)
the melting time 7. 215 hour for smooth tube, 5. 84 hour for
S team turbine; 6) Valve; 7) Condenser; 8) Pump. In the cold
dimpled tube, 5. 33 hour for cone fmned tube and 5. 04 hour for
hours or night cold HTF is pumped through TES unit which
helically fmned tube.
then extract heat from PCM . S o the temperature of outlet HTF
M. Bechiri and K. M ansouri [15] analyticaly simulated the
significantly increases to the inlet HTF.
LHTES system using shell and tube model. Exponential
integral function method and variable separation technique
A. Thermal Energy Storage Model

Proceedings of 2017 14th International Bhurban Conference on Applied Sciences & Technology (IBCAST) 582
Islamabad, Pakistan, 10th -14th January, 2017
 3

Fig. I: Design of Latent heat Thermal Energy Storage Unit for Solar Power Plant [22] [23].

1) Material Properties
High temperature PCM s are preferred for solar thermal power
plant applications. S o in this study inorganic PCM salts are The latent heat thermal energy storage (LHTES ) unit as shown
modelIed for the melting/solidification problem modelling of in Fig. 2 consist of cylinder to store PCM and tube for HTF.
TES system. The thermos-physical properties of the PCM s The working Heat Transfer Fluid chages/discharges the
salts used in this analysis are given in Table 1 and 2. It is PCM stored in the cylinder. Unsteady heat transfer from
assumed material is homogenous. circular tube and steady state interface evolution with time is
2) Thermal Energy StorageCylinder Dimension investigated by using following assumptions.
Fig. 2 shows the arbitrary dimensions of cylinder and tube I. The problem is axissymmetric (no angular variation)
dimensions. HTF flow passes through a tube of radius rland and two-dimensional.
transfer heat to the PCM stored in Cylinder of radius r2. This 11. Thermo-physical properties of HTF and PCM are
problem is congugate and HTF forced convection affects on constant, but properties of solid and liquid PCM is
the PCM melting rate is explored in this analysis. different as given in above Tables 1 and 2.

D,
III. Axial conduction in the fluid and solid material is
negligible.
IV. Flow of HTF in circular tube is Laminar with
HTF fiow Tube of
radius r, uniform heat flux condition.
V. The molten and solid PCM is homogenous and
isotropie.
Phase Change
Material
VI. As pressure in PCM is negligible so the phase change
of PCM occur at nearly constant temperature (i. e
KN03 at 330C)

I.
D,
.1
Fig. 2: Geometry for Phase Change Problem of TES system

Proceedings of 2017 14th International Bhurban Conference on Applied Sciences & Technology (IBCAST) 583
Islamabad, Pakistan, 10th -14th January, 2017
VII. The influence of natural convection on the melting of Thermal 0.5 0.51 4

PCM was studied by assuming effective thermal Conductivity(W/mK)

conductivity [15].
Viscosity - 1 .07

TABLE 1: Thermo-Physical properties ofKN03 Salt [20,21]

Properties Solid Liquid For the RTF (Reat Transfer Fluid):

r2
Melting Temperature ( 0c) 336 - u=2Um(l-- )
1J.2

Latent Heat (kJ/kg) 116 - 3Tf(z,r,t) 3Tf(z,r,t)

u + =
3x 3t
Density (kglm3) 21 1 0 21 1 0 (3)
32Tf(z,r,t) 13Tf(z,r,t)
(
3r- ? +-
r 3r )a =
Energy Density (kJ/m3) 244,760 -
3Tf(z,r,t)
--"----'-------=-- .... 0 S r S 1J.
Heat Capacity (J/mol-K) 0.3561 1 .39 3t
Tj (O,r,t) = 'Fa,z= O,t>0,0 r Ij (4)
Thermal 0.5 0.41 1
aTj (z,r,t)
= 0 r= O t>O
"
Conductivity(W/mK) ar (5)

Viscosity (mPa-s) - 2.09 Tf(z, r, t) = T; (z, r, t), r = lJ. (6)

1 00< Ra; < 1 07 For the Phase Change Material:

keif = O.386(
Pr )0.25 (Rac)025 Liquid Phase:

k O.861+Pr (1)
(ir; 1 or; 1 or;
--2 +--=--....1) rR(z,t)
Ra; or r or a, ot (7)
<100

k"'ff = 1
(2) (8)
k
S olid Phase:
III. GOVERNING EQUATIONS AND BOUNDARY
CONDITIONS
(9)
The mathematical model for two dimensional coordinate
a(z, r, t)
written in dimenssional form for RTF and PCM is written as = 0 ' r = r2 (10)
or
[15].
T,(z,r,t) =r;(z,r,t),t = 0,1) < r < r2
(11)
TABLE 2: Thermo-Physical properties of NaNOJ Salt [20,21] S olid/Liquid Interface condition:
( 12)
-k aT;(z, r, t) + k 8( z, r, t) - 8R -
Properties Solid Liquid
I
8r ., 8r
- pL
8t
, r R(z, t)
308 -
Melting Temperature (
T;( z, r, t) =(z, r, t) = T,1l, r = R(z, t) , t >0 (13)
C) In order to make this problem simple, less time consuming

Latent Heat (kJ/Kg) 1 99 - and general, we introduce dimensionless parameters. There

are the following benefits of introducing dimensionless
Density (kglm3) 2257 1 908
parameters [16].
Energy Density (kJ/m3) 449,1 43 -

I. It reduces the number of variables in equation.

Heat Capacity (J/kg/K) 3.1 445 1 .655

Proceedings of 2017 14th International Bhurban Conference on Applied Sciences & Technology (IBCAST) 584
Islamabad, Pakistan, 10th -14th January, 2017
 LI. Helps to analyze the system behavior without any r,
Where R2 --.2
-
consideration of units used to measure variables.
1]

iii. It helps to differentiate between relevant and

irrelevant variable and how it may affects the system. (23)
IV. It rescales variables and parameters so that we can (24)
have all the computed quantities of the same order.
(25)
v. It reduces the computation time during numerical and S olid-Liquid Interface:

analytical simulation. 8B(

, , Rpr) k, 8B/, R" r) _I 8R,
k,
=
, RJ S(';, r)
=
(26)
8R 8R, ste 8r
vi. Prevent round off errors.
Bs(';, Rpr) =
, ';, Rpr)
B( =
O, R, =
S(';, r), r > (27)
In view of these merits we introduce the following
dimensionless parameters in relation to the above equations.

For eed Conveetion

For the HTF:

Initial Co n dition
(15)
S(, r)

Where V(Rj) = 2(1- Rj2) (16)

BJ(O, Rpr) = I, = O, r > 0,0:s; RJ:s; I (17)
Inlet HTF
aB/ , Rpr)
----''--
-- = 0, RJ = 0,t > (18) Fig. 3: Initial and boundary conditions illustration in dimensionless form
aR,
BJ('RI'r) = B(
, , RI'r), R, =I
(19) These governing equations and boundary conditions describe
For the PCM: the congugate heat transfer problem of HTF and PCM in TES
Liquid Phase: system as shown in Fig. 3. In this analysis tube thickness at
a2B, + aB, aB, HTF and PCM is assumed to be negligible.
= 'I:S; RJ:s; S(J: r) (20)
aRJ2 RJ aR, ar , The solid and liquid temperature now can be expressed in the
aB,(, Rl'r) Nu
= (BJ ( J: R" r)-B) R, =1 form:
aR, 2K J , " (21)
B,(, Rpr) = B,(Rpr)x Bb() (28)
Bs(, R" r) = B(
S olid Phase:
s RJ, r) XBb() (29)
(22) At the interface of HTF-PCM we have

Proceedings of 2017 14th International Bhurban Conference on Applied Sciences & Technology (IBCAST) 585
Islamabad, Pakistan, 10th -14th January, 2017
 ae (,;)
(30) b = N2Ur(e(R ,r)-I)eb(J:
." ),R, =1 (41)
a,; "

(4 2)
S ince the flow inside the tube is laminar with unifonn surface
For the PCM:
temperature so average Nussult number can be calculated
Liquid Phase:

+ ae, = ael '1:::; R,

using this relation.
a2e,
:::; S(J:
.",r) (43)
aR,2 R, aR, ar
(31)
ael(R)'r)
= Bi(el(R)'r)-I),R, =1 (44)
With following conditions: aR
I
1
If [ ReD Pr/ (z / D) T3 ( af / at t 4 > 2: then (31) is used to
(45)

caIculate Nu value.
If 3 S olid Phase:
[ReDPr/(z/D)J/ (aJ/a,t4 <2:
then NUD = 3.66
Average convection coefficient is then caIculated using this
relation (46)
,; (z/D2 RePr)
(47)
=

3
NUD(';) =1.86(,;rIl (r tl4
(48)
aOj(!;,R"r) aOj(!;,R"r) TABLE 3: Dimensions of Cylinder
U(R,) + = (37)
a!; ar Diamater ofTube n (m) 0.32
aOj(!;,R"r (38)
L(R, ....O:5,R,:5,1
R, aR, aR, Diameter ofCylinder (m) 0.70

It is assumed forced convection heat transfer in a tube is Length ofCylinder (m) 1

arbitratry varied temperature by implying quasi steady state

assumption. Thus temperature distribution in working fluid is S olid-Liquid Interface:
independent of temperature distribution of heat transfer fluid l R)' r )
ae( 1 aR,
ks aes(R" r )
=_ R, = SC';, r) (4 9)
and depends upon the PCM temperature [17,18]. S o the bulk aR kl ste ar ,
aR,

1
es(R" r ) =e/R" r ) = O R, = SC';, r), r > 0
,
eb f RIU(Rl)ef(,Rl'r)dRl (50)
=
(3 9) IV. S olutions for the systems of Equatiow;
o

temperature of fluid given in equations (28) & (29) is given as For the RTF:
proposed by Bechiri [15]. The solution for (21) for the system of RTF is given as
Now differentiating(38) from 0 to linterval with respect to R1
and using equation (30) we get; (51)

(40)
(5 2)

Now subtituing (30) in (40) preveious systems of equations Equation (51) becomes:
can be written as: I;

For the RTF: eb() = exp(J a()d) (53)

o

Proceedings of 2017 14th International Bhurban Conference on Applied Sciences & Technology (IBCAST) 586
Islamabad, Pakistan, 10th -14th January, 2017
 Bix(Ei(-)-Ei(- R,2)
In conjugate problem of HTF and pe M (53) is used along
(58)
with temperature distribution of solid and liquid phase of e (RI' .)= 4. 4.
pe M as given in (28) and (29). Now in the next section we'lI
I
2exp(--) + Bix(Ei(--)
1 S2 -Ei(--)) 1
4. 4. 4.
determine solid and liquid phase temperature distribution.
According to (28) liquid phase temperature distribution of
pe M is written as:
1) Solution for Liquid Phase
The solution for (43) can be written by introducing Bix(Ei(- S2)-Ei(- Rl2) (59)
exponential integral function Ei (Rl) [19]. The solution for BI (c;, R" r) Bb ( 4r 4r
)
2exp(-_1 )+Bix(Ei(-2)-Ei(-_1 ))
=

liquid phase is now given as; 4r 4r 4r

(54)
The derivative of (54) with respect to dimensionless variable 2) Solution for Solid Phase
Rl is given as; The temperature distribution for solid phase of equations (46-
48) are solved by using the eigen function expansion method
(55) technique [16]. Solutions of these equations are written as:
Transcendental equation:
In order to [md the values of unknown variables A and B we
apply boundary codition (45), we get
Temperature distribution in solid phase:
A=BxEi(_R,2) (56)
4.
(61)
Now by putting value of A in (54) and then applying boundry
condition (44) we obtained the relation of second unknown :tn1 Cne(-r;r) (JO(An R1 )Ya(AnS) - JO(AnS)Ya(AnR1 ))
variable B which is given as;
m
B= 0n
2exp(-)+Bix(Ei(- S2)-Ei(-))
4. 4. 4.
By putting values of A and B in (54), temperature distribution
for liquid phase is

Fig. 4: Schematic of circular tube with specified boundary conditions

Proceedings of 2017 14th International Bhurban Conference on Applied Sciences & Technology (IBCAST) 587
Islamabad, Pakistan, 10th -14th January, 2017
Coefficient: RTF temperature at the melting temperature of PCM so t=tm
C" =
where tm is the melting duration of PCM. Equation (64) is
R,
solved using Mathematica notebook for both KN03 and
F(Rp r)( f (Jo(/t"R1)l'o(A"S)-Jo(A"S)l'o(A"R1)R1dR1) NaN03 salts. Solution approach for the melting duration is as
S(r)
R,
(62)
folIows:
f (Jo(A"RI)l'o(AnS)-Jo(AnS)l'o(AnRI))2 RdRI
S(r) a) For t=O RTF pipe is in contact with the PCM storage
Again according to (29) temperature distribution for solid cylinder and temperature of the heat source is
phase is modified into: equivalent to the melting temperature of PCM, so
Bs(,Rp r) = B(R!, r) = F(,Rl) = 1 at t=O.
(63)
b) Wall thickness of the RTF tube is assumed to be
00

Bb() L Cne(-r,;r) (Jo(A,I )l'o(AnS) -Jo(AnS)l'o(AnR1))

n1 negligible so the RTF and PCM are in contact to each
3) Interface Equation other.
Subtituing (61), (58) and (53) into (49) interface relation is c) Initial interface location is assumed to be at R1=1 and
obtained as given in the form given below it start to move toward R2 for t>O.
d) PCM melting is independent of RTF temperature
distribution by employing quasi steady state
2S exp( -1/ 4r) + Bi(Ei( _S2 / 4r) - Ei( -1/ 4r))

+ kI(;) ( Cne(-r,J,;r)( -J1(AnS(r))Ya(A"S(r)) assumption.

e) When the material is significantly melted then the
(64)
1 dS(r) natural convection effect are taken into account by
+AnJo(AnS(t))Y;(AnS(r)))
Ste introducing Rayleigh number given in equations (1)
= ---

dr

4) Solution for the Melting duration ofConjugate problem and (2).

As explained earlier this analysis is assumed by keeping the t) Transcendental equation given in (60) is obtained by
using window size method by assuming initial

1.8

1.7

1.6

c 1.5
0
:.;:::>
'Ci)
0 1.4
c..
<I>
<.> 1.3

2 1.2
c
- Axial position=1
--+- Axialposition=O.1
1.1
---.&--- Axial position=O.3
1.0 ---T- Axial position=O.7

0.9
0.0 0.5 1.0 1.5 2.0

Dimensionless time

Fig. 5: Effect of lnterface movement rate on dimensionless time at different axial positions for both salts at r =
0.4330467,
r =
1.216545, Ste=l.
Proceedings of 2017 14th International Bhurban Conference on Applied Sciences & Technology (IBCAST) 588
Islamabad, Pakistan, 10th -14th January, 2017
 interface location S=1 and outer cylinder boundary this evaluation natural convection effect was not assumed. In
R2=2. the next section effect of natural convection was studied on
g) Coefficient in (62) is nurnerically integrated in the the transition rate of PCM.
limits of S (R"t) =0 to 1. TABLE 4: Calculated thenno-Physical values forKN03 and NaN03 used
during computational calculations.
h) Calculated parameters values for KN03 and NaN03
given in Table 4 are used in the above equations for NaN03 KN03

analytical simulation of LHTES model. ks

as = - 0.5
= a' ===
pCp
pCp 7097.1 365 751.371

0.000705 0.0006655
IV. RESULTS AND DISCUSSIONS
a/=
k/
- = --- =
0.51 4 a,= -
k,
= --
0.411 =
1) Melting rate and Axial Temperature Distribution pCp 31 57.74
pc" 2932.9

In order to find the interface location of the melted PCM, (64) 0.0001 628
0.0001401
a, 0 0006655 =
Y=-=
is solved using Mathematica. Equation was solved by as
Y= - =
0.000705
= a, 0.0001401
a/ 0.0001 628
assuming the constant value of Stefan nurnber (=1). 4.7501784
0.4330467
And results were plotted against function of dimensionless
cPCIII (I0 T111 )
time denoted as t. It was noted KN03 takes more time to meh cpcm(To-T,I/) = Ste =
-

Ste=
L L
completely than NaN03 which is due to higher 1 .1 O(To - T"J
0.95(r;, - T,,,)
116
melting/transition temperature of KN03 than NaN03. Analysis 1 99 = 0.0081897(r;, - T,,,)
= 0.0055276(To-T,I/)
of transition rate at different axial position indicated, increase
melting rate was achieved at lower axial position as shown in ks ks 0.5
0.9727626 -=--=

k, 0.514 0.411
= =

k,
Fig. 6. PCM close to the inlet HTF has higher heat transfer
1.216545
than the outlet side of PCM. This is due to the fact HTF
deposit its heat to the material during charging process that's
why material locating close to the inlet HTF melts abruptly as 2) EiJect ofNatural Convection
compare to the other. Material melting rate is not homogenous To study the natural convection effect on the melting rate
in large cylindrical system that's why in order to increase the effective thermal conductivity given in (l) is introduced in the
melting rate homogenously fins heat exchanger are employed solid/liquid interface equation. Buoyancy driven forces act
to gain the uniformity and better heat transfer rate in literature under the g force shows the convection effect known as
[2]. natural convection. Natural convection effects become
Temperature distributions for PCM along radial and axial significant when considerable fraction of PCM salt is melted.
regions are shown in fig 7. In order to validate the analysis In order to study the natural convection effect it is important
temperature distribution at t=O is evaluated by using initial to find the value of Prandtl number. Prandtl number for the
condition given in equation 4.3. As equation 4.3 suggest cases of KN03 and NaN03 are calculated using the defined
temperature of PCM is a function of axial position and time, values given in Table 1 and 2. Prandtl nurnber is a
and we assumed F (, R, ) =1, this can also be confirmed by dimensionless parameter used in the study of heat transfer
analysis between solid body (cylinder) and moving fluid
Fig. 7. Temperature distribution in PCM varies along the axial
(mohen sah). It is defmed as the ration between thermal and
and radial direction. PCM transition rate decelerated with
momentum diffusivity, written in simplified form of (20);
increasing axial position due to the heat losses to the PCM
close to the nearer heat source as explained above. During

Proceedings of 2017 14th International Bhurban Conference on Applied Sciences & Technology (IBCAST) 589
Islamabad, Pakistan, 10th -14th January, 2017
 z
2.0

1.8

1.6

0.5 1.4

1.2

0.0 1.0
1.0
1.5
2.0
0.0 0.5 1.0 1.5 2.0
R

R
1.0

R
0.16
1.0

2.0
___ =<-] 0.00
0.12

0.05

0.08
0.10

0.15 O.O

0.20

0.0 z

Fig. 6: 3D representation of temperature distribution of axial and radial position at (a) t=0 (b) t=1 (c) t=2.5 for peM (r =0.433, Ste=l)

Proceedings of 2017 14th International Bhurban Conference on Applied Sciences & Technology (IBCAST) 590
Islamabad, Pakistan, 10th -14th January, 2017
 time of PCM. So in order to increase the storage capacity it is
Pr= CpclIIfl (65)
k,
--

required to provide more energy and time to the PCM to store

Using relation (65) calculated values for KN03 and NaN03are heat. That's why cylinder radial length also affects the storage
7.06 and 3.44 respectively. These values are used in (1) to find capacity of TES unit. If the radial length is too small PCM
the effective thennal conductivity in order to investigate the melting rate will be achieved too soon and most of heat will be
convection effect on the melting of PCM. The important lost to the ambient or outer cylindrical boundary. So in order
parameter for the calculation is the Rayleigh number which is to achieve better energy storage, TES system should be
associated with buoyancy driven flow. Rayleigh number for designed as to give enough time to melt the whole PCM
the case of heat transfer in radial tube direction is given as: within the required period during charging process.
1. 7
Ra= g CFa -Tn)(R2 -R,)3 ...(66)
via,
1.6
C
0
Where is the thermal expansion coefficient and v is the .;::;
. 1.5
</l
0
kinematics viscosity of molten salt. By using above relation p.., 1.4
<l)
u
Rayleigh number for NaN03and KN03 are calculated 1.3
<t<l)
12
56257.175 and 33469.945 respectively by assuming .....

..s
l.l
(1'0 -7;,) =l. Now by putting values of Rayleigh and Prandtl
number in (1), effective thermal conductivity for NaN03 and
1.
0 t;.-------::--
- -================::J 4

Dimensionless time
KN03 has been calculated which is equal to 5.58 and 5.06
Fig. 7: Effect of natural convection on the melting rate at for NaN03 (
respectively. r = 0.433, Ste=O.15, =I)
Fig. 7 and 8 shows the effect of natural convection on the
melting of KN03 and NaN03salts. These figures c1early
shows natural convection c1early accelerate the heat transfer
rate in the case of NaN03, but it does show any significant
effect on the transition rate of KN03 and transition rate of
KN03 was also reported to slow as compare to NaN03. This
was due to the fact that viscosity of molten KN03 salt was
higher than the viscosity of NaN03, which affected the
buoyancy driven forces due to the higher flow resistance of 1.
0 L;-------;-----=========::::!J 10
molten salto Dimensionless time

3) Radial Temperature Distribution

In Fig. 6(a), (b) and (c) it can be seen PCM start melting from
Fig. 8: Effect of natural convection on the melting rate at for
inner radius of cylinder R,=1 and ends at R2=2. Initially at KN03 (r 4.75, Ste=O.II, =1)
=

(t=O) PCM is assumed to be in the solid phase at the

temperature equal to the transition temperature of PCM as CONCLUSION
V.

shown in the contour plot of PCM at t=O. As the time passes A mathematical model for high temperature phase change
temperature distribution increases in the radial direction of material based thermal energy storage system was studied to
cylinder until the whole PCM is melted at R2=r2/rj=2. Energy predict the thermal behavior of cylinder and tube storage unit.
storage capacity of the TES unit depends upon the charging A several calculations were perfonned to study the effects of

Proceedings of 2017 14th International Bhurban Conference on Applied Sciences & Technology (IBCAST) 591
Islamabad, Pakistan, 10th -14th January, 2017
different pertinent and geometrie parameters. The obtained [7] Lane GA. Slar heat strage: latent heat materials, vl. T. Flrida:
CRC Press, Tnc.; 1983.
analytieal solutions gave a good estimation of thermal [8] Lane GA. Slar heat strage: latent heat materials. Technlgy,

behavior and parametrie effeet of PCM whieh ean be used as [9]

vl.ll. Flrida: CRC Press; 1986.
Dincer I, Rsen MA. Thermal energy strage, systems and
referenee for experimental or numerieal data validation. applicatins. Chichester (England): Jhn Wiley & Sns; 2002.
[10] Bellecci, c., and M. Cnti. "Transient behaviur analysis f a
REFERENCES latent heat thermal strage mdule." Internatinal jurnal f heat
[I] Dincer, Ibrahim. "Evaluatin and selectin f energy strage and mass transfer 36.15, 3851-3857, 1993.
systems fr slar thermal applicatins." Internatinal Jurnal f [ll] Zhang, Yuwen, and Amir Faghri. "Semi-analytical slutin f
Energy Research 23.12, 1017-1028, 1999 thermal energy strage system with cnjugate laminar frced
[2] Sharma A, Tyagi VV, Chen CR, Buddhi D. "Review n thermal cnvectin." Internatinal jurnal f heat and mass transfer 39.4,
energy strage with phase change materials and applicatins." 717-724, 1996.
Renewable and Sustainable energy reviews 13.2, 318-345, 2009 [12] Esen, Mehmet, Aydin Durmu, and Ayla Durmu. "Gemetric
[3] Ip K, Gates 1. Thermal strage fr sustainable dwellings. In: design f slar-aided latent heat stre depending n varius
Proceedings f internatinal cnference sustainable building, parameters and phase change materials." Slar energy 62.1, 19-28,
Maastricht, The Netherlands;2000. 1998.
[4] Paris, Jean, Michel Falardeau, and Cecile Villeneuve. "Thermal [13] Esen, Mehmet. "Thermal perfrmance f a slar-aided latent heat
strage by latent heat: a viable ptin fr energy cnservatin in stre used fr space heating by heat pump." Slar energy 69.1, 15-
buildings." Energy Surces 15.1, 85-93, 1993. 25, 2000.
[5] Mehling, Harald, and Luisa F. Cabeza. Heat and cld strage with [14] Tao., Y. B., Y. L. He, and Z. G. Qu. "Nurnerical study n
PCM. Berlin: Springer, 2008. perfrmance f rnlten salt phase change thermal energy strage
[6] Sharma, R. K., et al. "Develpments in rganic slid-liquid phase system with enhanced tubes." Slar Energy 86.5, 1155-1163, 2012.
change materials and their applicatins in thermal energy strage." [15] Bechiri, Mhammed, and Kacem Mansuri. "Analytical slutin
Energy Cnversin and Management 95,193-228, 2015. f heat transfer in a shell-and-tube latent thermal energy strage
system." Renewable Energy 74, 825-838, 2015

Proceedings of 2017 14th International Bhurban Conference on Applied Sciences & Technology (IBCAST) 592
Islamabad, Pakistan, 10th -14th January, 2017
[16] Khalid M.Z, Zubair M. and Ali M, "A novel method for analytical
solution of transient heat conduction and Stefan problem in
cylindrical coordinate," 2016 13th International Bhurban
Conference on Applied Sciences and Technology (IBCAST),
Islamabad, 2016, pp. 476-482.
[17] Cao, Yrn, and A. Faghri. "Performance characteristics of a thermal
energy storage module: a transient PCM/forced convection
conjugate analysis." International Journal of Heat and Mass
Transfer 34.1, 93-101, 1991.
[18] Cao, Y., and A. Faghri. "A study of thermal energy storage
systems with conjugate turbulent forced convection." Journal of
Heat Transfer 114.4, 1019-1027, 1992.
[19] M.N.Ozisik, Heat conduction. 2nd ed. New York: John Wiley and
Sons; 1980, ASIN: B0086HWUI4.
[20] Takahashi, Y., R. Sakamoto, and M. Kamimoto. "Heat capacities
and latent heats of LiN03, NaN03, and KN03." International
Journal of Thermophysics 9.6, 1081-1090, 1988.
[21] Agyenim, Francis, et al. "A review of materials, heat transfer and
phase change problem formulation for latent heat thermal energy
storage systems (LHTESS)." Renewable and sustainable energy
reviews 14.2, 615-628, 2010.
[22] Hoshi, Akira, et al. "Screening of high melting point phase change
materials (PCM) in solar thermal concentrating technology based
on CLFR." Solar Energy 79.3 (2005): 332-339.
[23] Gumaste, R.A., 2011. Computational simulations of latent heat
thermal energy storage systems-with innovative and first
principles based simulation for the underlying unsteady melting
(and solidification) processes.

Proceedings of 2017 14th International Bhurban Conference on Applied Sciences & Technology (IBCAST) 593
Islamabad, Pakistan, 10th -14th January, 2017
 Modeling Of Common Cause Failures (CCFs) by
using Beta Factor Parametric Model
Qazi Muhammad Nouman Amjad, Muhammad Zubair Gyunyoung Heo
Department Of Basic Sciences Department of Nuclear Engineering, Kyung Hee
University Of Engineering & Technology, Taxila. University, South Korea.
[email protected]
Abstract (PSA), especially within PSA fault tree (FT) modeling of
safety systems in nuclear power plants. The issue of CCF
Nuclear accidents and incidents such as Three Mile Island (TMI- has attracted a substantial academic attention, through years
2) accident (1979), Chernobyl disaster (1986) and the recent as well as lately.
Fukushima nuclear disaster (2011) have caused people to be
This paper presents a convenient method for explicit
suspicious of the safety of nuclear energy, and have reduced the
level of trust among public. Common cause failure (CCF) has been
modeling of single component failure event simultaneously
a major element of such accidents in terrestrial nuclear power within several different common cause failure groups
reactors because of high redundancy built into the systems and (CCFGs). Each CCFG is dened on the basis of specic
susceptibility of these redundant systems to CCF mechanisms. For coupling mechanism. The presented methodology that
this purpose, ad hoc approaches used to be taken to address accommodates single component failure event to be
vulnerabilities to CCF by operating staff of the plants. A CCF simultaneously assigned to different CCFGs given different
event is a result of simultaneous failure of two or more individual coupling mechanisms is based on a modication of the
components. Such an event can signicantly affect the availability frequently used and to the scientic community well-known
of safety systems and has long been recognized as an important
Beta Factor parametric model. It is the most commonly used
issue in the probabilistic safety assessment (PSA). So a
complicated and unresolved problem in the subject of safety and
CCF-model, and was originally proposed by Fleming [1].
reliability is to model CCF in PSA. To overcome this problem the This model assumes that a certain percentage of all failures
present research highlights a mathematical model to estimate are CCFs. When using the beta-factor model, we have to
system unavailability in nuclear power plants (NPPs) as well as in assume that each element of a system can fail in two
other industries. This mathematical model is based on Beta Factor different ways: as an independent failure that only affects
parametric model. The motivation for development of this model the element considered, or as a dependent failure (CCF)
lays in the fact that one of the most widespread software such as where all the elements in the subsystem fail at the same time
for fault tree (FT) and event tree (ET) modeling as part of the PSA (or within a short time interval).The motivation for this
does not comprise the option for simultaneous assignment of single
study is the incapability of one of the most widespread PSA
failure event to multiple CCF groups.
software for FT and event tree (ET) modeling [2], for
A signicant nding from such modeling is that, in contrast to simultaneous assignment of one single component failure
common expectations, a too early nuclear phase-out will not serve event in more than one CCFG within the fault tree analysis
the deployment of renewable energy sources and rational use of (FTA) technique.
energy. The proposed method can be seen as an advantage of the
explicit modeling of CCF. 2. Modeling of common cause failures in the
probabilistic safety assessment
1. Introduction 2.1. Overview of PSA
PSA is being acknowledged as the most effective tool for
The common cause failures (CCFs) refer to a specic class safety and risk management in NPPs. The two most
of dependent failure events that are considered to have a commonly used techniques for system modeling within PSA
potential of simultaneous occurrence due to a shared cause. for NPPs are the fault tree analysis (FTA) and the event tree
This shared cause is an implication of a simultaneous analysis (ETA). The purpose of system modeling in PSA is
existence of a root cause and a coupling mechanism. The to provide an abstract representation of the ways in which
root cause is identied as the most basic cause of systems can fail to perform their intended functions FTA
component failure which, if corrected, would prevent and the ways in which system successes and failures interact
reoccurrence of the cause. The coupling mechanism with one another in the course of accident sequence
implicates the condition for multiple components to be progressions ETA [3].
affected by the same cause. In general, the susceptibility of FTA is a tool to identify and assess all combinations of
a system containing redundant components to dependent undesired events in the context of system operation and its
failures, as opposed to independent failures, is determined environment that can lead to the undesired state of a system.
by the presence of the coupling mechanisms. Undesired state of the system is represented by a top event.
CCFs are being acknowledged as one of the most Logical gates connect the basic events (BEs) to the top
challenging issues in the probabilistic safety assessment event. BEs are the ultimate parts of the FTA, representing
University of Engineering and Technology, Taxila, Pakistan.
Kyung Hee University, South Korea.
different undesired events such as component failures, Where P( BEi ) is the probability of occurrence of basic
missed actuation signals, human errors and common cause
event i within the considered minimal cut set. Thus, the
contributions [4,5]. The qualitative phase of the analysis
solves the system logic function in terms of the minimal cut following simplied expression of (1), can be derived on
sets (MCSs), which are combinations of the smallest the basis of assumption of rare event approximation and
mutual independence of occurrence among BEs, (2),:
number of component faults that may cause the system
N xn
fault. Each of the MCSs is calculated as a Boolean logical
product of basic events. The quantitative phase calculates PTOP = P( BE
n =1 i =1
ni ) (2)
the probability of system failure, i.e. the fault tree top event
probability of occurrence PTOP : Where P( BEni ) represents the probability of occurrence of
PTOP = basic event i comprised within minimal cut set n.
N 2.2. Incorporation of CCF in the FTA
PMCSn ( BE1 ,......,BExn ) _
n =1
There are basically two ways to incorporate dependent
failures in PSA and in any system reliability analysis in
general: implicit and explicit methods. The implicit
P
n<m
MCS n ( BE1 ,......,BExn ) MCS n ( BE1 ,......, BExi ......)
+
approach models the failure causes, being included in the
PSA model on a higher level, more complexly. The implicit
approach suggests that the minimal cut sets and the Boolean
probability equation are rst considered and presented as if
P
n<m
MCS n ( BE1 ,......, BExn ) MCS n ( BE1 ,......, BExi ......) all basic events are mutually independent basic events.
Mutual independence among basic events in general means
that the joint probabilities are equal to the products of
 MCSl ( BE ,......, BEx ...)..... + individual basic events probabilities of occurrence. After
1 i
N Boolean reduction is conducted, the terms of the algebraic
(1) N 1 P MCS n ( BE ,......, BEx probability expression are quantied in a way that the CCF
n)
n =1
1
contribution is subsequently added to the top event
probability, independently of whether this top event
(1) probability stands for system reliability or availability [6].
An example is given on Fig. 1, where the top event, e.g. a
Where MCSn is the minimal cut set n, N is the number of system failure, occurs if all three mutually dependent
MCSs, PMCSn ( BE1 .... BE X ) is the probability of occurrence of BEs: E1 , E2 and E3 representing probabilities of failure of
the homonymous components ,occur simultaneously
nth MCS ( MCS n ) which is comprised of xn basic events. (logical AND gate). Since mutual independence among BEs
For a well-maintained and well-managed systems such as in general means that the joint probabilities are equal to the
NPP safety systems, the probability of component failure is products of individual BE probabilities of occurrence, in
very small in comparison of the probability of the each term of the derived Boolean equation, any product
component being in operational state, which approaches P ( BEi ) P( BE j ) is then replaced with
unity for a given time interval. These kinds of systems are
also known as rare event systems. One important feature of P ( B E i B E j ) , P ( BEi ) P ( BE j ) P ( BE x ) with
the rare event systems is that in the process of calculating its
failure probability one can often omit higher order terms, P ( BEi BE j BEx ) etc. However, the implicit app-
such as the second and third term on the right-hand side of roach is constrained with the assumption that the joint
(1), without substantial accuracy loss. probabilities P ( BEi BE j BEx .....) are known or can
Additionally, most of the computer codes [2] assume the
BEs to be mutually independent in the process of be determined through correlations or conditional
probabilities [7, 6, and 8]. Different parametric models are
quantication. In such case, the probability of nth MCS is associated to the implicit approach. On the other side, the
calculated as follows: explicit approach is appropriate when the CCF cause is
evident and may be included in the PSA or any other kind of
PMCSn (BE1....BEX ) =
xn

P (BEi ) (2)
reliability model as a separate basic event or functional
block. The explicit approach is especially appropriate and
i =1
 of this explicit approach, are valid also for components E2
andE3 , i.e.
E2 = E2 ind = E2 ind ; E3 = E3ind C13 = E3ind
P ( E2 ) = P( E2 _ ind ) and
P ( E3 ) = P ( E3 _ ind ) + P(C13 )
Although the number of BEs and MCSs is smaller than in
the explicit method, the implicit method has some
disadvantages [6, 8].
useful for consideration of intersystem and inter-component In case of non-computerized quantication process the large
dependences [9]. Certain types of dependent events can be number of MCSs may need to be manually quantiable.
modeled in a FTA explicitly as mutually independent basic Additionally, there is a risk of losing terms, which could be
events Cij ... for failures of specic components i,j , Such important through the mutual dependencies they impose
among each other, due to the applied truncation in the
a basic event Cij is dened as the probability of course of the manual quantication process.
simultaneous presence of components failed states, which is 3. Method
represented by the corresponding BEs and indicated by the
3.1. Method description
sub-indices i and j, i.e. BEi and BE j . In general, each
The parametric models can be divided into two major
component- level FT basic event BEi , representing failure categories: shock and non-shock models. While the shock
models take into account the failure mechanisms, the non-
of component i, is replaced with the union of BEi ind and shock models estimate BEs probabilities without postulating
all events Ci ... whose occurrence would mean occurrence a model for the underlying failure mechanisms. The non-
shock group of parametric models comprises the Basic
of the corresponding BEi as well. BEi ind is a basic event Parameter model, the Beta Factor (BF) model, the Alpha
that represents the portion of the total probability of Factor model, the Square Root model, the Multiple Greek
occurrence of BEi which is due to independent causes. The Letter (MGL) model, the Multiple Beta Factor (MBF)
model. The Binominal Failure Rate Model (BFR) is
probabilities of theseCi ... events are dened in a way that accounted as shock model. The objective of all these
parametric models is to develop the probability of the
all joint probabilities P ( BEi ) , P ( BEi BE j ) are
occurrence of BEs on a set of parameters. Although each of
retained, thus gaining correct values for all the MCSs as the models is characterized with certain specics, all of
well as for the fault tree top event probability PTOP . This them share the same general idea. This common idea is the
step is equivalent to a redenition of the logic model basic fact that component failure probability, i.e. each
events from a component-level basis to a deeper components failure space, can be divided to a dependent and
resolution. This deeper resolution of consideration provides independent portion. The relation between both portions is
with the capability for explicit modeling of CCFs within determined with different parameters dened within the
FTA. In other words, it is due to this deeper level of detail specic parametric model selected [10].
that identies the effects which common cause events of This paper presents a modied Beta Factor parametric
specied multiplicity may pose on the system. The explicit model as a prerequisite for the herein proposed method for
method does not depend on the system success criterion and simultaneous assignment of single component failure event
is not constrained by the number of BEs per MCS. For within multiple CCFGs. The standard Beta Factor
parametric model is a non-shock, single parametric model,
example, if components E1 and E3 have been assigned to i.e. it uses one parameter in addition to the total component
a common cause failure group CCFG1 , the BEs in the FT failure probability to calculate the CCF probabilities.
According to the Beta Factor Parametric model, a fraction
on Fig. 1 are expanded to include the BE C13 , dened as a beta of the component failure rate or component failure
simultaneous failure of E1 and E3 due to a common cause. probability can be associated with common cause events
shared by the other components in that group. Another
From this chapter onwards, the above-described symbolic assumption is that whenever a common cause component
Cij ... will be replaced with CCFGX referencing the failure mode occurs, all components within the
homonymous CCFG which the simultaneously failed common cause component group are assumed to fail [11].
components i, j, belong to. Fig. 2 shows this expansion 3.2 Example
for the case of BE E1 . Analogous FTA expansions, as part In order to present briey how the Beta Factor parametric
 Namely, W different may exist in theory, i.e.
1 ,..., w . Additionally, it is often assumed that the
probabilities of similar events, involving similar types of
components, are the same. The proposed method takes
advantage of the physical symmetries associated with
identically redundant components.
Thus, the following expression, equivalent to the logic given
by Fig. 4, is valid:

model works, let us assume an example system of two

redundant pumps (Fig. 3). Both of the pumps have to fail in
order the system fails. If the pumps are mutually
independent, then the failure probability of the system P
(sys) (modeled as FT top event TOP on Fig.3) would be:
2
P( sys ) = PTOP = P( P1 ) P( P2 ) = PP1 PP 2 = Pcomp (3)

P1 = P1_ ind CCFG1 CCFG2 ... CCFGw (5)

Where PP1 and PP 2 are probabilities of valves failures. i.e.:
Redundant component identicality assumption implicates:
P( P1 ) = PP1 = [Independent portion] + [CCF portion]
PP1 = PP 2 = Pcomp. In case CCF exists between the pumps =
and standard Beta Factor model is utilized, each of the BEs P( P1_ ind ) + P(CCFG1 ) + P(CCFG2 ) + ... + P( CCFGw )
P1 and P2 (Fig. 3) will be substituted with a sub-  
  

independenportion CCFportion
component FT as on Fig. 2. Thus, system failure probability (6)
would be assessed as follows: In such way, component P failure space is divided into
PTOP = (1 ) Pcomp (1 ) Pcomp + Pcomp (4) dependent, CCF-affected portion:
 
 

CCF portion =
P1 _ ind P 2 _ ind P2 _ CCF
P(CCFG1 ) + P(CCFG2 ) + ... + P(CCFGw ) = PCCFG1
Where PCCF = . PP1 =. PP2 =. Pcomp herein.
+ PCCFG2 ++ PCCFGw =
Now, let us assume that there are U physically identical
pumps per train, connected in series, and let pump P be W
coupled with other pumps from train B within W different 1 PP1 + 2 PP1 + ... + w PP1 = PP1 x
CCFGs due to W different common cause coupling x =1
mechanisms. Let the number of components within each of Where: PCCFG1 = 1 PP1 ; PCCFG2 = 2 PP1 ; ..
the CCFGs be arbitrary. Let all U pumps from train B be
encompassed by the CCFGs with arbitrary distribution . PCCFGw = w PP1 (7)
throughout the CCFGs. Analogous to the example on Fig. 1, and independent portion:
Each of BEs in the component-level FT modeling failures of
Independent portion = P ( P1_ ind ) = PP1_ ind , such as :
individual pumps, which are being comprised within
individual CCFGw ( w = 1,..., W ) , will be additionally W

modeled, i.e. expanded with a sub-component-level FT as ( PP1_ ind + PP1 x ) = PP1 < = > PP1_ ind
= (1 1 2 , ...,
1 w ) PP 1 (9)
on Fig. 2. Specically, BE P1 , modeling failure of pump
Thus, above equations dene the convenient method for
P1 , will be expanded i.e. replaced with the following FT: simultaneous assignment of single failure event into several
different CCFGs, where as Equation (9) specically
describes the proposed Modied Beta Factor model as a
complementary part of the presented method.
The implications on system level, after application of the
presented methodology and assigning P1 simultaneously to
W different CCFG, are more or less trivial, i.e.:
PTOP = [CCF contribution] + [independent contribution]
U U
= [ PCCFG
1
+ PCCFG2 + ... + PCCFGw ] + [ PPiTRA TRA
_ ind Pj _ ind ] (10)
i =1 j =1
Fig. 5 depicts the equivalent FTA for calculating probability new method presented in this paper, a complementary
of considered system failure. approach for equalizing the cumulative CCF portion (in case
CCF portion breaches the total component failure space) to
TOP
the total component failure space is presented.
(CCF)
First, a proportional reduction factor (PRF) is introduced:
W
PRF = 1/ (
x =1
X ) (9)

after which a new set of reduced beta factors is introduced:

1' = 1 PRF ; '2 = 2 PRF ;..., 3' = 3 PRF ;
W W
CCFG2
(
Independent CCFG1 CCFG3 '
Portion = > x ) = PRF[ x ]=1 (10)
x =1 x =1
PCCFG1 PCCFG2 PCCFG3
Fig. 5. Equivalent FT for the scenario with W different CCFGs In this way the logical inconsistency, i.e. CCF portion
. breaching component P1 total failure space, will be, at least
3.3. Method limitations and an approach to resist against from theoretical aspect, diminished. Given this PRF-based
The beta-factor model has several weaknesses. When using approach to cope against the eventual logical inconsistencies
the beta-factor model, it is not possible to distinguish that may arise in the course of applying the proposed new
between a CCF involving all elements and a multiple failure method for explicit modeling of single component failure in
involving two or three elements [12]. The proposed method
allows a more detailed consideration of common cause
failures. Due to lack of data, the beta-factor is often
estimated by checklists, such as those given in Humphreys
[13], IEC 61508 [14], Johnston [15], and Brand [16].
When developing PSA system models applying FTA, in
which CCFs are not considered, the basic events
representing failures of different components are accounted
as independent. As it was mentioned earlier in this paper, Fig. 6. Partition of component failure space .
the introduction of CCFs within the analysis implicates a more different CCFGs, new proportion between the
partition of the components failure space to an independent independent and CCF portions within the component failure
and dependent part in terms of causes. The independent part space is implicated:
is the portion of the components failure space which is [CCF portion ] =
'
result of an independent cause. The dependent part is the
portion of the components failure space which occurs due to P (CCFG1' ) + P (CCFG2' ) + ... + P(CCFGw' ) =
a common cause coupling mechanism that couples the
considered component with other components within a
PCCFG' + PCCFG' + ... + PCCFG' =
1 2 w
specic CCFG, which are being affected by the same W
coupling mechanism. 1' PP1 + '2 PP1 + ... + 'w PP1 = PP1 'x (11)
If we consider, for example, the case from the previous x =1
subsection where component P1 is being simultaneously And:
'
assigned to W different CCFG (Fig. 4) due to W different [Independent portion ] = P( P1_' ind ) = PP' 1_ ind = [1-
coupling mechanisms, then component P1 total failure W

space, P( P1 )= PP1 , is dened with Equations. (6), (7). Let

x =1
'
x ]=0 (14)

us suppose a situation in which, e.g. due to selection of beta Consequently, there will be implications on system level
factors based on an expert judgment, calculations as well. The following equation presents the
(1 + 2 + 3 +,..., +w ) > 1 . In such a case, the CCF newly calculated probability of failure for the specic
example system analyzed in 3.1. (Figs. 3 to 6):
portion of the considered component ( P1 ) failure space
PTOP = [CCF portion ]' + [Independent portion ]'
practically becomes bigger than the total component failure
space. In order to cope with logical inconsistencies of this =[ PCCFG' + PCCFG' + ... + PCCFG' ] +
1 2 w
kind (Fig.6), a general reduction of all beta factors U U U
(12)
separately in a way that the CCF portion becomes less than P (P '
) +
P 1 _ ind
TR B
j _ ind
'
P TR A
P 1 _ ind ( PPTjR_Bind ) '
or, at most, equal to the total component failure space needs j =1
 
i=2 j =1

to be implemented. At this stage of the development of the =0

The approach presented in this subchapter is one of the [3] Haasl D, Young J, Cramond WR. Probabilistic risk assessment course
documentation. NUREG/CR-4350, vol.4. US NRC, Washington, DC;
options for constraining the possible mentioned logical
1985.
inconsistencies. Other approaches for constraining the [4] Cepin M. Optimization of safety equipment improves safety. Reliability
possible logical inconsistencies, which may arise, are Engineering and System Safety 2002;77:7180.
possible as well. In that sense, one such approach that can [5] Roberts NH, VeselyWE, Haasl DF, Goldberg FF. Fault tree hand-book.
NUREG-0492. US NRC, Washington, DC; 1981.
be considered is equalizing the cumulative CCF portion (in
[6] Vaurio JK. An implicit method for incorporating common-cause
case CCF portion breaches the total component failure failures in system analysis. IEEE Transactions on Reliability 1998;47:173
space) to the largest one among the W beta factors. In both 80.
of these two coping approaches, the former, applied within [7] Mosleh A, Rasmuson DM, Marshall FM. Guidelines on modeling
common- cause failures in PRA.
the manuscript and the latter one, briey described with the
[8] Vaurio JK. Common cause failure probabilities in standby safety
previous sentence, the probability of occurrence of each of system fault tree analysis with testing-scheme and timing dependencies.
the CCFGs comprising the CCF portion is just a product of Reliability Engineering and System Safety 2003;79:4357.
the component total probability and the corresponding beta 9] Zio E. Computational methods for reliability and risk analysis. Series on
quality, reliability and engineering statistics, vol. 14. Singapore: World
factor, i.e. there is no weighting of probability of occurrence
Scientic Publishing Co. Pte. Ltd; 2009.
of different CCFGs ((8),). In that sense, other general 10] Cepin M. Application of common cause analysis for assessment of
possibility that can be studied and investigated from the reliability of power systems. In: Proceedings of 2010 IEEE PMAPS
aspect of more detailed and comprehensive analysis is e.g. Conference, Singapore; 1417 June 2010.
[11] Mosleh A, Parry GW, Paula HM, Worledge DH, Rasmuson DM.
weighting of the probabilities of occurrence of the different
Procedures for treating common cause failures in safety and reliability
CCFGs comprising the CCF portion. This weighting directs studies. NUREG/CR- 4780, Vol.2. US NRC, Washington, DC; 1989.
assignment of W weighting factors w f i , such that NUREG/CR-5485. US NRC, Washington, DC; 1998.
[12] Hokstad P, Rausand M. Common cause failure modeling: status and
W trends In: Misra K, editor. Handbook of performability engineering.
w
i =1
fi = 1 : The rationale for the selection of the weighting London,
[13] Humphreys RA. Assigning a numerical value to the beta factor for
common cause evaluation. In: Proceedings of the sixth conference,
factors should be based on expert judgment or engineering Birmingham, UK; 1987. p. 2C/5/18.
experience. [14]IEC 61508. Functional safety of electrical/electronic/programmable
As expected, there are quantitative and qualitative electronic safety-related systems, 2nd ed. Geneva: International
implications on the results in cases where there is need to Electrotechnical Commission; 2010.
[15] Johnston B. A structured procedure for dependent failure analysis.
apply the PRF-approach. Namely, the shares of the MCSs Reliability Engineering 1987;19:12536.
representing CCF portion of specic components, subjected [16] Brand P. A pragmatic approach to dependent failures assessment for
to the PRF-approach, are decreasing in general. Implicitly, standard systems. AEA Technology PLC; 1996.
the shares of the MCSs representing the corresponding
independent portions are increasing. However, the top event
probability experiences negligible quantitative change.
4. Conclusions
CCFs contribution is an important parameter that strongly
inuences the reliability of the considered systems. If CCFs
are considered, the system failure probability increases. So
this paper presents a convenient method to model CCF in
PSA using beta factor model that acts as a strong defence
against CCF, thus mitigating the incidents & the accidents.
So, it will help to remove the doubts of the people about
safety of nuclear energy and to rehabilitate the level of trust
among public. The results and insights implicate improved,
more transparent and more comprehensive models within
PSA.
Acknowledgment
This work has been supported by NSSC project, Kyung Hee
University, South Korea and University of Engineering and
Technology, Taxila, Pakistan.
References
[1] Fleming KN. A reliability model for common mode failures in
redundant safety systems. Report GA-A13284, General Atomic Company,
San Diego, CA; 1975.
[2] Relcon AB. Risk Spectrum PSA. Sweden; 1998.
 Annals of Nuclear Energy 90 (2016) 106114

Contents lists available at ScienceDirect

Annals of Nuclear Energy

journal homepage: www.elsevier.com/locate/anucene

Calculation and updating of Common Cause Failure unavailability

by using alpha factor model
Muhammad Zubair a,, Qazi Muhammad Nouman Amjad b
a
Nuclear Engineering Department, University of Sharjah, 27272, United Arab Emirates
b
Department of Basic Sciences, University of Engineering and Technology, Taxila, Pakistan

a r t i c l e i n f o a b s t r a c t

Article history: The most lethal role of Common Cause Failures (CCFs), which motivate the experts to investigate it, is the
Received 23 May 2015 dependent behavior therein contained, which leads to simultaneous failure of the systems. Highly redun-
Received in revised form 3 December 2015 dant systems are more susceptible to be affected by CCFs and also CCFs have been recognized as the prin-
Accepted 8 December 2015
cipal contributor in the terrestrial reactor accidents. In the past, plenty of work has been done regarding
the calculation of unavailability of different types of systems due to CCFs by using different techniques
such as fault tree analysis (FTA). But the qualitative aspects such as human errors, maintenance faults
Keywords:
and poor components quality cannot be updated by using FTA as the changes occur. So in order to over-
Reliability
Failure probability
come this problem, multinomial distribution function and its conjugate Dirichlet distribution function
Common Cause Failures has been used as likelihood and prior, respectively, in Bayes theorem to obtain an updated posterior func-
Auxiliary Feed Water System tion of the same form as Dirichlet distribution function thus improving the working and monitoring capa-
bility of Probabilistic Safety Assessment (PSA). Furthermore, the presented research highlights a
mathematical model to estimate system unavailability due to CCF by using alpha factor model. By using
this model, we can calculate failure probability (unavailability) of the systems quite accurately through
the two parameters ak and Qt. The ease of using the proposed model can be assessed through the brief
analysis of a case study of Auxiliary Feed Water System (AFWS). AFWS is used in all designs of
Pressurized Water Reactor (PWR). It plays a vital role in maintaining a heat sink by providing feedwater
to the steam generators.
2015 Elsevier Ltd. All rights reserved.

1. Introduction
Components fail within a selected period of time such that

success of the PRA mission would be uncertain.
A CCF event occurs (NUREG/CR 5485, 1998) when;
Component failures result from a single shared cause and
coupling mechanism.

Two or more individual components fail or are degraded,
A component failure occurs within the established component
involving failures during demand, in-service testing, or deficien- boundary.
cies that would have resulted in a failure if a demand signal had
been received. The Common Cause Failures (CCFs) has been under considera-
tion since the time of the reactor safety study wash 1400
(NUREG/75-014, 1975). CCFs belong to a particular class of depen-
Abbreviations: PSA, Probabilistic Safety Assessment; CCF, Common Cause dent failure events that are considered to have a potential of simul-
Failure; CCCG, common cause component group; BE, basic event; PWR, Pressur- taneous occurrence due to a shared cause. This shared cause is an
ized Water Reactor; AFWS, Auxiliary Feed Water System; MDPs, motor driven implication of a simultaneous existence of a root cause (most basic
pumps; TDPs, Turbine Driven Pumps; CST, Condensate Storage Tank; MOVs, Motor cause of component failure which, if corrected would prevent reoc-
Operated Valve; AOVs, air operated valves; MOOS-M, maintenance out of service
for motors; MOOS-T, maintenance out of service for turbines; FTS-M, failure to start
currence of the cause, it tells Why did the item fail?) and a coupling
for motors; FTS-T, failure to start for turbines; FTS-D, failure to start for drivers; mechanism (implicates condition for multiple components to be
FTR-M, failure to run for motors; FTR-T, failure to run for turbines; FTO, failure to affected by the same cause, it tells Why were several items
open; FTC, failure to close; FRC, failure to remain closed. affected?). Coupling mechanism is the property that makes multi-
Corresponding author. Tel.: +971 567146701.
ple items susceptible to the same root cause. Fig. 1 elaborates the
E-mail address: [email protected] (M. Zubair).

http://dx.doi.org/10.1016/j.anucene.2015.12.004
0306-4549/ 2015 Elsevier Ltd. All rights reserved.
 M. Zubair, Q.M.N. Amjad / Annals of Nuclear Energy 90 (2016) 106114 107

E1

E2

E3

Fig. 1. Shared causes.

Fig. 2. Common cause analysis steps.

above mentioned theory of shared cause, where Ei denotes that the Alpha factor model is commonly used model and a strong Baye-
ith component is in a failed state. sian approach to this model has recently been proposed by
Modeling of CCFs is being acknowledged as one of the most (Troffaes et al., 2014) using a set of conjugate priors instead of a
demanding issues in the Probabilistic Safety Assessment (PSA), single prior distribution, following a standard approach for gener-
especially within PSA fault tree (FT) modeling of safety systems alized Bayesian methods in theory of imprecise probabilities. The
in NPPs. This challenging issue of CCF has attracted a substantial alpha-factor parameterization of the basic parameter model starts
academic attention, through years as well as lately. This paper pre- out with considering the total failure rate of a component which
sents a conducive method for the modeling of CCFs to calculate the could involve failure of any number of components, that is, this
system failure probability by using alpha factors. Each Common is the rate obtained by looking at just one component, ignoring
Cause Failure Group (CCFG) is defined on the basis of particular everything else. For the present study, the parametric alpha factor
coupling mechanism. The presented methodology that accommo- model is chosen because the alpha factor model can handle com-
dates single component failure event to be simultaneously mon cause component group sizes of different levels; can be
assigned to different CCFGs given different coupling mechanisms adopted even when no statistical data on Common Cause Failure
is based on an alpha factor parametric model. rates are available; and is more accurate compared to other para-
PSA is being acknowledged as the most effective tool for safety metric models. The alpha factor model estimates the CCF frequen-
and risk management in NPPs. The two most commonly used tech- cies from a set of ratios of failures and the total component failure
niques for system modeling within PSA for NPPs are the fault tree rate. The parameters of the model are Q t and ak , where Q t is the
analysis (FTA) and the event tree analysis (ETA). The purpose of total failure frequency of each item due to all dependent (CCFs)
system modeling in PSA is to provide an abstract representation and independent events and ak is the fraction of total failure events
of the ways in which systems can fail to perform their intended frequency and include k components failure due to a common
functions FTA and the ways in which system successes and failures cause.
interact with one another in the course of accident sequence The frequency of common cause basic event involving k compo-
progressions ETA (NUREG/CR 4350, 1985). nents in failed state in a system of m components is given by:

1
k m1 ak Q t
2. Alpha factor model For a staggered testing scheme : Q m
 1
k1

The parametric models consist of two major categories: shock

k a
models and non-shock models. Those models which take into For a non-staggered testing scheme : Q m
 k Qt
k m1 2
account the failure mechanisms to estimate BEs probabilities are k1
at
termed as shock models whereas, those models which estimate P
where at = m k1 kak . System unavailability for a staggered testing
BEs probabilities without taking into account the failure mecha-
scheme for a three component system can be written as:
nisms are termed as non shock models. The basic parameter model,
the Beta Factor (BF) model, the alpha factor model, the Square Root 3
model, the Multiple Greek Letter (MGL) model, the Multiple Beta Q s 3a1 Q t 2 a2 Q t 3a3 Q t 3
2
Factor (MBF) model are the examples of the non-shock group of
parametric models whereas the Binominal Failure Rate Model Whereas, system unavailability for a non-staggered testing scheme
(BFR) is an example of a shock model. All of these parametric mod- for a three component system can be written as:
els tend to evaluate the probability of the occurrence of BEs on a a3 a2 a3 a2 a3
Qs 3 Q 3 QT 3 QT 4
set of parameters. Although each of the models is recognized with at T at at
particular parameters, all of them share the same general idea. This
general idea is the fact that component failure probability, i.e. each
components failure space, can be separated into two parts: a 3. A brief overview of CCF analysis
dependent portion and an independent portion. The relation
between these two portions is determined with different parame- This example presentation involves some procedural steps
ters defined within the specific parametric model selected. (NUREG/CR 5485, 1998).
108 M. Zubair, Q.M.N. Amjad / Annals of Nuclear Energy 90 (2016) 106114

Fig. 3. Steps for problem definition and system modeling.

The steps for problem definition and system modeling are prior and specific data is used to calculate the posterior (Zubair
shown in Fig. 3. et al., 2010, 2014)
This section includes the definition of analysis objectives,
boundary conditions, mission time, potential system alignments, 4.1. Multinomial likelihood function and its conjugate Dirichlet
environmental hazards (like earthquake), basic events and the rel- distribution
evant component failure modes, potential operator actions, and
any other assumption for analysis. Distribution of the alpha factor model parameters: the estima-
CCF vulnerabilities and corresponding component group tors for the alpha factor model parameters for redundant compo-
responsible for that vulnerability are identified by using both nents systems are given by:
quantitative and qualitative approaches. n
The objective of qualitative approach is to determine the events ak Pm k 5
k1 nk
that are the principal contributors to system vulnerabilities by
focusing on failure mechanisms and ignoring plant specific The data must be of the following form: E = {nk : k = 1, . . ., m} to
defenses. The determination of these events based on judgment estimate a0k s. The likelihood of observing this evidence, given a
and feedback from operating experience, to ensure that all applica- bunch of values for a0k s is:
ble events are included and those events that are not applicable are
excluded from the analysis. Again for convenience, well deal in Cn1 n2    nm Y m
Ln1 ; n2 ; . . . ; nm ja1 ; a2 ; . . . ; am ank 6
primary events so that the numbers of events remain at a manage- Cn1 Cn2 . . . Cnm k1 k
able level i.e. too much conservatism is ignored in this analysis. At
his stage, analyst must decide which groups of components have a In this methodology the likelihood distribution function is given by
significant likelihood of experiencing a common cause event multinomial distribution function. Multinomial distribution func-
affecting two or more components within that group. It is neces- tion shows likelihood of possible outcomes of an experiment with
sary to understand the factors shown in Fig. 4 that will be formally repeated trials in which each trial can result in a specified number
considered in root cause analysis (see Fig. 5). of outcomes that is greater than 2. It is based on the assumption
The motivation for studying quantitative approach is the inca- that a0k s provide the probability of the n0k s subject to the constraint
P
pability of one of the most widespread PSA software for FT and that the sum of a0k s is one (i.e. m
k1 ak = 1). The most common appli-
event tree (ET) modeling (Relcon, 1998), for simultaneous assign- cation of multinomial likelihood function in PSA is the utilization of
ment of one single component failure event in more than one CCFG the alpha factor model for the CCFs.
within the fault tree analysis (FTA) technique. FTA is a tool to iden- Dirichlet distribution function can be expressed as follows:
tify and assess all combinations of undesired events in the context
of system operation and its environment that can lead to the unde- CYk Y
k

sired state of a system. Undesired state of the system is repre- f prior aY1
i 7
CY i1
k
sented by a top event. Logical gates connect the basic events
(BEs) to the top event. BEs are the ultimate parts of the FTA, repre- where Y = sti : Here s is greater than zero and t i belongs to (k  1)
senting different undesired events such as component failures, P
dimensional unit simplex i.e. ki1 t i 1
missed actuation signals, human errors and common cause contri-
The above equation can also be written as:
butions (Cepin, 2002; NUREG/CR 0492, 1981) (see Fig. 6).
It is usually desirable to decrease the size of the problem under Y
k
investigation. For this purpose, quantitative analysis is often used f prior / : aiY1 8
which is in turn also known as quantitative screening analysis. In i1

other words, we can say that the objective of quantitative approach

These distributions are multivariate extensions of the Con-
involves the truncation of those common cause component groups
strained Non Informative (CNI) priors presented in (Atwood,
which have not any significant affect on the total system unavail-
1996). A motivating example is the alpha-factor model for
ability (see Fig. 7).
common-cause failures, described below. If we use multinomial
This truncation is very significant for system reliability analysis,
distribution function Eq. (6) as likelihood and Dirichlet distribution
sometimes becomes necessary for accident-level analysis and it is
function Eq. (7) as a prior in Bayesian network, a posterior distribu-
achieved by expanding component level fault tree i.e. by inserting
tion function which is also Dirichlet in form is obtained.
a global common cause basic event for every CCFG in the
According to Bayes theorem the posterior distribution is related
component level FT.
to the prior distribution by:
f post / Prx=pf prior 9

4. Methodology Substitute values from Eqs. (6) and (7) in Bayes theorem (Eq. (9)):
Cn1 n2 nk Qk ni CYk Qk
Cn1 Cn2 Cnk a
i1 i CYk i1 aY1
i
There are two types of data (Generic data and specific data) f p=k R 1 Cn1 n2 nk Qk ni CYk Qk
which are required to update unavailability parameters. Generic 0 Cn1 Cn2 Cnk
a
i1 i CYk i1 aY1
i dai
data are easily available commercially (Shafaghi, 2008) whereas,
specific data are collected at facility (e.g. nuclear plant reliability After canceling the constants from numerator and denominator;
data system) (NUREG/CR 2300, 1983). Fig. 2 shows that these data Qk ni Qk Y1
a a
do not provide any assistance to evaluate the reliability of the plant fp=k R 1 Qki1 ni Qki1 i
therefore Bayes theorem is used in which generic data serve as 0
a
i1 i
i
a
i1 i
Y1
dai
 M. Zubair, Q.M.N. Amjad / Annals of Nuclear Energy 90 (2016) 106114 109

Root Cause
Analysis

Whether groups of Consideration of

identical potential for
components are human errors

Extent of diversity Physical proximity Susceptibilities of

among components of redundant components to
in a redundant components varied

Fig. 4. Prominent factors that must be included in root cause analysis.

Incorporation Of global Reduce Obtain MCS Evaluate

common cause basic Problem size Unavailability
event

Incorporation of Solve FT to Calculate CCF basic

FT basic events obtain MCS event probabilities

Fig. 5. Quantitative analysis steps.

After solving this equation the posterior distribution which is The marginal distribution of ak is a beta distribution with mean and
also Dirichlet in form is obtained which is given by: mode given by:
Y
k Mean:
fpost / aY1n i
10
i
A nk Ak
i1 ak Pm 0k ; k 1; 2; . . . ; m 16
i1 A0i ni A T
From above equation it can be deducted explicitly as:
Mode:
fpost Y  1prior ni 11
A n0k  1 Ak  1
Multinomial distribution and its conjugate Dirichlet distribu- ak;mode Pm k ; k 1; 2; . . . ; m 17
i1 A 0i ni  1 A T 1
tion has used in Bayesian theory to update number of observations
of each failure that acts as a strong defense against CCF. If we know For a uniform prior Ak = 1 where k = 1, . . ., m.
prior parameter to observe the data point, then we can easily Let we discuss an example to understand how these obtained
update our knowledge based on data point and finally end up with results serve to update the number of observations of each failures.
a new distribution of the same form as the old one thats where the The probability of k out of m components given that failure has
distributions we used are prominent because they are conjugate happened, is denoted by ak. The prior expectation for each ak is
prior of each other. Thus we can successively update our knowl- lspec,k (Atwood, 1996; Nathan, 1998).
edge of parameter by incorporating new observations one at a Number of components = 4; lspec,1 = 0.950; lspec,2 = 0.030;
time, without running into mathematical difficulties. lspec,3 = 0.015; lspec,4 = 0.005
Another way to express our result from Eq. (7) is given below: Number of observations = 36.
CA1 A2    Ak A1 1 A2 1 35 observations showed one component failing and one obser-
pa1 ; a2 ; . . . ; ak a a2    aAk k 1 12
CA1 CA2 . . . CAk 1 vation showed two components failure.
Therefore, n1 = 35; n2 = 1; n3 = 0; n4 = 0
where A0sk are the parameters of posterior distribution and are
tk = lspec,k and s = 21l1
related to a similar group of prior distribution parameters spec;k

[A01 ; . . . ; A0m ] through following relationship: Using all the given data in the formulas of tk and s, we can cal-
culate Y, since Y = stk.
Ak A0k nk 13 So, prior values are Y  1 = 8.5; whereas the updated posterior
where 0 6 Ak 6 18k values are Y  1 + nk = 43.5 for k = 1.
The general form of marginal distribution of each ak is a beta
distribution of the form: 5. Case study
CAT
pj aj aAJ 1 1  aj AT Aj 1 14
CAj CAT  Aj j The system chosen as a case study is a three train Auxiliary Feed
Water System (AFWS) which is basically belongs to the category of
where Pressurized Water Reactor (PWR). AFWS has eleven designs classes
X
m depending upon the number of pumps and steam generators.
AT Aj 15 Design class 10 contains a Condensate Storage Tank (CST), three
j1 pump trains (two motor driven pumps (MDP-1 and MDP-2) and
110 M. Zubair, Q.M.N. Amjad / Annals of Nuclear Energy 90 (2016) 106114

closed. Each of the MDPs can supply flow to two dedicated steam
generators through MOVs assuming that valves successfully open
on demand and steam TDP can supply up to all of the four steam
generators under the same assumption (NUREG/CR 5485, 1998).
AFWS provides feed water to the steam generator in case of any
emergency resulting in a loss of main feed water. Usually, the loss
of main feed water occurs due to the following reasons: (a) Loss of
Coolant Accident (LOCA), or (b) loss of offsite power, or (c) pumps
shutdown, or (d) low water level in CST. The three train pumps of
AFWS can begin their functioning in case of any emergency either
manually or automatically. MDPs are started automatically when-
ever offsite power is lost or main feed water pumps shut off or low
water level is detected whereas TDP is started automatically when-
ever low water level is detected or offsite power is lost. One of the
very prominent features of these three train pumps is that all the
pumps are mechanically identical in spite of the employment of
diversity in these pumps. This important feature of these pumps
makes this system susceptible to the CCF mechanism. The system
success is defined as the CST flow through two MDPs or one TDP to
two of the four steam generators. Valves function to allow or regu-
late flow to steam generators for the success of the system.
The analysis objective is to determine AFW system unavailabil-
ity and to determine principal contributors to the system failure.
Both point estimates and interval estimations are provided
through probabilistic approach quantification. For convenience,
only primary components (CST, MDPs, TDPs, MOVs, actuation cir-
cuitry) are considered for investigation. The boundaries of the pri-
mary components are described in the Table 1. Note that for
convenience only the prominent boundary conditions are consid-
ered, other boundary conditions like electric power supply or
external events (like fire) are not included in this analysis.
The example under our consideration has three natural groups
of components: (1) four identical MOVs, (2) three identical pump,
and (3) two identical motor drivers. It also includes check valves
but we exclude it from our analysis for convenience. Furthermore,
it has been a common observation of operators since long that
common cause events rarely affect the diverse components there-
Fig. 6. Methodology to obtain updated posterior distribution function. fore, the components which are not in the selected groups are
assumed to be diverse i.e. no CCFs will be considered for these
one Turbine Driven Pump (TDP), four steam generators and four components.
Motor Operated Valves (MOVs). This design class has been selected Algebraic equation for system unavailability denoted by U:
for investigation out of eleven design classes because most of the
nuclear power plants in the world contain the AFWS with design U S U C U VG U PG U MG U P U T 4U V 3 U V 3
class 10. The pumps take suction from a common CST and supply
U P 2 2U M U T 4U V 2U M U P U T U M 2 U P U T
header to accommodate AFW flow to four steam generators. The
four Motor Operated Valves at pump discharge remain normally 4U V U P U T 4U V U M U P 4U V U M U T 18

Fig. 7. Major components in AFWS.

 M. Zubair, Q.M.N. Amjad / Annals of Nuclear Energy 90 (2016) 106114 111

Table 1 From the 1987 to 1995 experience, it was observed that the TDPs
Boundary conditions of some of the primary components. unreliability was greater than the MDPs unreliability and DDP
Component name Boundary conditions unreliability by a factor of 10 and by a factor of 4 respectively thus
Motor Operated Motor/operator, operator circuit breaker, indication making TDPs least reliable (NUREG/CR 5500, 1998).
Valves (MOVs) circuit, control circuit and switches, torque limit Some values of alpha factors of different components of AFWS
bypass switch, and valve hardware based on 19871985 experience are listed in Table 3. Here a2=2
Pumps Motor and turbine driver, circuit breaker, control and a3=3 of MDPs or TDPs represent the values of alpha factors
circuit and switch, pump hardware
when 2 out of 2 and 3 out of 3 MDPs or TDPs failed to start. In
the third row of above table a2=2 , a3=3 and a4=4 represent the values
of alpha factors when 2 out of 2, 3 out of 3 and 4 out of 4 pumps
Alpha factors calculated from the 19871995 experience in NUREG/
excluding driver failed to run. Similarly, in the fourth row a2=2 ,
CR 5500 (1998)) and the alpha factors estimated in NUREG/CR 5497
a3=3 and a4=4 represent the values of alpha factors when 2 out of
(1998) (which are usually taken as standard) have been given in
2, 3 out of 3 and 4 out of 4 feed segment flow control valves failed
Tables 2 and 3 respectively.
to open. In the last row, a2=2 represent the values of alpha factor
The industry wide arithmetic mean of AFWS unreliability in
Probabilistic Risk Assessment (PRA) i.e. standard value for a mis- when 2 out of 2 steam supply valves to turbine failed to open. In
sion time of 24 h is 0.00034. Whereas, the arithmetic mean of the following graphs (Fig. 8) the values of alpha factors of MDPs,
AFWS unreliability calculated from the 1987 to 1995 experience TDPs and valves are plotted on the basis of above mentioned data
is 0.0021. Furthermore, the arithmetic mean of MDP, TDP and Die- in Table 3, through which the identification of the prominent CCF
sel Driven Pump (DDP) is 0.0025, 0.022 and 0.0057 respectively. contributor becomes quite easy.
Values of alpha factors
Values of alpha factors

Fig. 8. A comparison between estimates of alpha factors based on the 19871995 experience and as mentioned in literature.

Table 2
Alpha factors of some important components of AFWS for n = 2, 3 and 4 as estimated in (NUREG/CR-5497).

Components Failure mode NC a1=2 a2=2 a1=3 a2=3 a3=3 a1=4 a2=4 a3=4 a4=4
AFWS FTS 22 .9213 .0787 .9226 .0311 .0463
MDPs FTR 19 .9766 .0234 .9556 .0351 .0093
TDPs FTR 23 .9762 .0238 .9704 .0171 .0125
AOVs FTC 5 .9693 .0307 .9585 .0290 .0125 .9496 .0349 .0056 .0100
AOVs FTO 7 .9677 .0323 .9534 .0359 .0108 .9530 .0250 .0148 .0072
MOVs FTC 11 .9677 .0323 .9471 .0394 .0134 .9430 .0320 .0157 .0094
MOVs FRC 4 .9823 .0177 .9803 .0098 .0100 .9777 .0098 .0044 .0081
AFWS FTS 19 .9657 .0343 .9626 .0192 .0183
112 M. Zubair, Q.M.N. Amjad / Annals of Nuclear Energy 90 (2016) 106114

Table 3 Cumulative Probability

Estimates of alpha factors based on 19871995 experience in NUREG/CR-5500.

Components Failure mode a2=2 a3=3 a4=4 0.5

MDPs FTS .028 .016

TDPs FTS .068 .0022 0.4
Pumps: excludes divers FTR .012 .0031 .0019

Cumulative Probability
Feed segment flow control valve FTO .018 .0026 .0014
Steam supply valves to turbine FTO .085 0.3

0.2

Table 4
AFWS failure mode data. 0.1

Failure mode Unplanned Independent

demands failure probability 0.0

Failures Demands
0.0000 0.0001 0.0002 0.0003 0.0004 0.0005
MOOS-M 4 1995 0.195
Unavailability
MOOS-T 5 602 0.175
FTS-M 6 1993 0.160
Fig. 11. Cumulative probability distribution of the total system unavailability.
FTS-T 16 597 0.010
FTS-D 1 65 0.370
FTR-M 1 1987 0.368
FTR-T 2 583 0.271 6. Results and discussions
Total FTS probability for motors 10 1993 0.125
FT recover for motors, CCF 1 2 0.25
events
The estimates calculated by alpha factor methods are lower
Total FTR probability for pumps 1 2635 0.368 than the estimates derived directly from the experience. The alpha
Failure to recover MOOS-M 2 4 0.375 factor method was chosen over the direct method for several rea-
Failure to recover MOOS-T 3 5 0.345 sons. First, the data used for the calculating the CCF estimates
directly is limited to only lethal events. That is, only lethal CCF
events found in the unplanned demand data are used in the direct
calculation which ignores the effects of partial system failures due
By using the data given in Tables 24, one can easily calculate to CCF. The CCF Data Collection and Analysis system contains par-
AFWS components unavailability (QS). Independent failures proba- tial event information thereby providing a richer source of infor-
bility (Qt) and alpha factors are used in Eqs. (3) or (4) depending mation for evaluating and quantifying CCF. Secondly, the CCF
upon the testing mechanism to calculate total system unreliability. estimates computed directly did not differentiate with regard to
No. of Failures

Fig. 9. Causes of unplanned demand failures by failure mode.

No. Of Failures

Fig. 10. Causes of segments failures.

 M. Zubair, Q.M.N. Amjad / Annals of Nuclear Energy 90 (2016) 106114 113

5000
Probability density function

4000

Probability density function

3000

2000

1000

0.0000 0.0002 0.0004 0.0006 0.0008 0.0010

Unavailability

Fig. 12. Probability distribution of the total system unavailability.

the common cause group size thereby providing an estimate with- Common Cause Failure terms. The failure event in which all the
out regard for group size. As a result, applying a single estimate to a three pumps and four MOVs failed to perform their intended func-
system with different levels of redundancy is not appropriate tions is the most dangerous and probable CCF event. The results
(NUREG/CR 5500, 1998). Therefore, based on these reasons we declare that the dependent failures i.e. CCFs contribute over 93%
can say that the direct estimates are conservative as compared to of the total system unavailability whereas the remaining percent-
the alpha factors estimates. age of failures is caused by independent failures terms.
The use of more than one component/train in AFWS increases
the overall reliability of the system because in case where one train
7. Conclusions
fails to perform the intended safety function, the other train does it.
However, it was observed on almost nine occasions that more than
It can be easily observed that the calculated average value of
one train fail to perform its intended safety function. CCF, failure of
AFWS unreliability from 1987 to 1995 experience is greater than
common suction source and error of commission were the major
the average values in PRA by a factor of six. This implies that there
causes of those failures. The graph shown in Fig. 9 represents the
are some factors like human error, poor maintenance quality,
main causes of some failures of TDP and MDP. It is a very convenient
design deficiencies etc which are responsible for the increase in
way to identify the principle contributor through the above men-
unavailability of the systems making the systems unreliable. So,
tioned graph. It shows that the main cause of failure to start and
the situation demands for the improvement in CCF defenses,
failure to run of TDP was observed to be hardware failure (trip link-
strategies, monitoring and updating. This paper presents a conve-
age either being out of adjustment, worn excessively, or stem bind-
nient method to model CCF in PSA by using alpha factor model.
ing, bearing failure etc). Maintenance involves failure to start event
The most common application of the multinomial likelihood in
in MDP. Other errors also contribute minutely in FTS and FTR events
PSA is the alpha factor model for CCF, which is used as a stochastic
in TDPs and MDPs. Error of commission i.e. FTR is caused by opera-
model for dependent failures on demand of redundant components
tor intentionally action. Suction Segment failure involves some of
due to a shared cause i.e. CCF, was presented. In this application,
the failures occurred following a reactor trip caused by loss of steam
the parameters of interest are the multinomial probabilities of
generator level control. The Common Cause Failures linked with
the failure of k redundant components, ak, and the conjugate prior
feed control segments were caused by two hardware problems,
is the Dirichlet distribution with parameters Qk In this application
an environmental problem, a personnel error, and a maintenance
failures of more than one component are rare, and so the marginal
error. Water accumulation failures were classified as FTS and these
mean of a1 is near to unity, making Q1 the dominant parameter in
failures resulted in a turbine overspeed trip. The design category of
the Dirichlet prior. The results and insights implicate improved,
failures involves both FTR and FTS. The support system failures
more transparent and more comprehensive models within PSA.
were considered as outside the system boundaries study.
So, all of the above mentioned defenses against CCFs will increase
There were many failures of feed control segments as shown in
the reliability of the considered systems and they will help to
Fig. 10. Hardware failures contributed almost more than half of the
remove the doubts of the people about safety of nuclear energy
total failures, while maintenance and personnel errors accounted
and to rehabilitate the level of trust among the members of the
for approximately 12% and 18% of the failures, respectively. Design
public.
and environmental-related problems accounted for the few
remaining failures. The following graphs in Figs. 11 and 12 repre-
sent the relation between total system unavailability and their Acknowledgment
cumulative probability. In both graphs, the total system unavail-
ability is taken on x-axis whereas the cumulative probability and The authors are thankful to University of Sharjah, UAE.
probability density function are taken on y-axis respectively. The
uncertainty distribution for the total system unavailability is repre- References
sented by these graphs (NUREG/CR 5485, 1998). The main charac-
teristic of the distribution is the mean value that is equal to 0.0002. Atwood, C.L., 1996. Constrained non informative priors in risk assessment. Reliab.
Eng. Syst. Safety 53, 3746.
Now, with all the above mentioned results studied, calculated Cepin, M., 2002. Optimization of safety equipment improves safety. Reliab. Eng.
and plotted on graphs show that the failures are dominated by Syst. Safety 77, 7180.
114 M. Zubair, Q.M.N. Amjad / Annals of Nuclear Energy 90 (2016) 106114

Nathan, O.S., Kelly, Dl., 1998. Bayesian parameter estimation in probabilistic risk USNRC, NUREG/CR-5500, 1998. Auxiliary/Emergency Feed water System Reliability,
assessment. Reliab. Eng. Syst. Safety 62, 89116. 19871995, INEEL/EXT-97-007401987, vol. 1, pp. 2260.
USNRC, NUREG-75/014 (WASH-1400), 1975. Reactor safety study: an assessment of Relcon, A.B., 1998. Risk Spectrum PSA, Sweden.
accident risk in U.S. commercial nuclear power plants, pp. 102110. Shafaghi, A., 2008. Equipment failure rate updatingBayesian estimation. Int. J.
USNRC, NUREG/CR- 4350, 1985. Probabilistic risk assessment course Hazard. Mater. 159, 8791.
documentation, vol. 4, Washington, D.C. Troffaes, M.C.M. et al., 2014. A robust Bayesian approach to modeling epistemic
USNRC, NUREG/CR-5485, 1998. Guidelines on Modeling Common-Cause Failures in uncertainty in common-cause failure models. Reliab. Eng. Syst. Safety 125, 13
Probabilistic Risk Assessment, INEEL/EXT-97-01327, pp. 79109. 21.
USNRC, NUREG/CR-0492, Fault tree hand-book, Washington, D.C., 1981. Zubair, M. et al., 2010. Calculation and updating of reliability parameters in
USNRC, NUREG/CR-2300, 1983. A guide to the performance of probabilistic risk probabilistic safety assessment. J. Fusion Energy 30 (1), 1315.
assessment for nuclear power plants, In: Human Reliability Analysis, vol. 1, pp. Zubair, M. et al., 2014. Quantitative and qualitative analysis of safety parameters in
100140. nuclear power plants. Int. J. Energy Res. 38 (6), 755764.
USNRC, NUREG/CR-5497, 1998. Common Cause Failure Parameter Estimation,
INEEL/EXT-97-01328.