ADALYA JOURNAL https://doi.org/10.37896/aj10.
8/020 ISSN NO: 1301-2746
Estimation of Mean Time between Failures of a Repairable System
With Common Cause Shock Failures
Changkuoth Jock Chol 1 , Bang Chuol Nhial 2 and Dr. G. Y. Sagar 3*
1, 2 Lecturer, Department of Statistics, College of Natural and Computational Sciences, Gambella University, Ethiopia.
3 Professor, Department of Statistics, College of Natural and Computational Sciences, Gambella University, Ethiopia.
Abstract
Present paper discusses method of maximum likelihood (M L) estimation to evaluate the estimates of mean time
between failures (MTBF) of a system. A two unit identical system in series and parallel subject to individual, lethal
and non-lethal common cause shock (LCCS & NCCS) failures is studied. Life times and repair times of the units
are exponentially distributed. We derived M L estimates of mean time between failures functions in both series and
parallel systems in the absence of analytical approach. We have established the validity of the derived estimates
through simulation process empirically.
Key words: M L Estimation, Mean Time between Failures, Markov Process, Common Cause Shock failures,
Simulation
Introduction
The enumeration of Common Cause Shock (CCS) failures is of importance because these induce simultaneous
failure in the system and drastically reduce reliability of the system. The common cause events may be outside
environment such as fire, thunderstorm, lightning, flood, earthquake, error due to human interventions etc. Occur at
random times causing simultaneous failure of multiple components. The reliability analysts and researchers were studied
two types of CCS failures in literature. (i) Lethal common cause shocks results failure of all the units in the system and (ii)
A non-lethal common cause shocks results failure of some units at random. Reliability analysts were discussed them in the
evaluation of reliability measures and routine of the system very much. The CCS failure models can be found in the
literature by Billinton and Allan (1983). Some other reliability models with CCS failures were analyzed by Chari et al
(1991), Verma (1991), Sagar et al (2007).
Separately from mathematical modelling, estimation and life testing is vital interest in order to measure average
life of the device etc. In reality life testing experiments are intended to measure the average life of the units and also
fascinated to counter such questions as ‘what is the possibility that the system will down in the interval (0, t). Awgichew et
al (2018) derived M L estimates of availability measures for two unit system with CCS failures and also established
simulation study on reliability estimates of a repairable system. Sreedhar et al (2012) proposed maximum likelihood
estimation approach for estimating reliability indices of a two unit system in the presence of CCS failures. Levitin (2001)
proposed the universal generating function of multi-state system dependability analysis to include CCS failures. A
procedure is suggested to evaluate reliability functions of no repairable series-parallel multistate systems under the
influence of CCS failures. Alexopoulos and Shultes (2001) discussed the method of estimating reliability expressions in
extremely dependable Markov systems. Reddy (2003) developed reliability expressions under the influence of lethal and
non-lethal common cause shock failures of a two unit non-identical system.
The purpose of present investigation is to incorporate M L estimation approach in Sagar et al (2019). We evaluate
the M L estimates of mean time between failure functions for series system as well as parallel system. The empirical
approach is used with Monte Carlo Simulation.
Assumptions
We consider a two unit identical system with the following assumptions.
i) The units fail individually and also simultaneously due to two kinds of CCS (lethal or non lethal) failures in Poisson
fashion.
ii) Individual, LCCS and NCCS failures are independent to each other.
iii) There is a single repair facility to repair failed units whether they are failed individually or simultaneously due to
common cause shock.
iv) Repair times of failed units depend on the failure mode and are assumed exponentially distributed.
Volume 10, Issue 8, August 2021 189 http://adalyajournal.com/
ADALYA JOURNAL https://doi.org/10.37896/aj10.8/020 ISSN NO: 1301-2746
Notations
: Individual failure rate
: Rate of LCCS failure
: Rate of NCCS failure
1 : Repair rate when one unit is down and other one is working
2 : Repair rate when second has failed whereas one unit was already down
c : Repair rate when both units fail simultaneously
p ( q ) : The probability of simultaneous failures of units due to NCCS (LCCS)
RLNS (t ) : Reliability of the series system in (0, t) with LCCS and NCCS failures
RLNP (t ) : Reliability of the parallel system in (0, t) with LCCS and NCCS failures
ELNS (T ) : MTBF of the system with LCCS and NCCS failures when units are in series
ELNP (T ) : MTBF of parallel system in case of LCCS and NCCS failures
Eˆ (T ) : M L Estimate of MTBF function for series system under the influence of LCCS & NCCS failure mode
LNS
Eˆ LNP (T ) : Estimate of MTBF function for parallel system in the presence of LCCS and NCCS failures
x , y , w : Sample means of individual, NCCS and LCCS failure occurrence
z : Average repair times
xˆ , yˆ , wˆ : Estimates for individual, NCCS and LCCS failure rates of occurrence respectively
ẑ
: Estimate of repair times
n : Size of sample
N : Simulated samples
M S E : Mean Square Error
Markov Model
Under the stated assumptions Markov model can be formulated to derive the mean time between failure function
E (T) under the influence of individual as well as CCS failures and the state transition diagram is given in Figure.1. The
numerals in Figure.1 denote the system state.
λc
1 – (λ1+λc)
1 – (λ2+µ1) 1 – (µ2+µc )
λ1 λ2
0 1 2
µ1 µ2
µc
Figure.1 State transition diagram
The quantities in the above model are as follows:
1 2( pq )
2 ( p 2 )
c ( p )
Volume 10, Issue 8, August 2021 190 http://adalyajournal.com/
ADALYA JOURNAL https://doi.org/10.37896/aj10.8/020 ISSN NO: 1301-2746
Mathematical Modelling
Based on the arguments of stochastic theory, we can develop the set of differential equations associated with the
existing mathematical model for the above mentioned state transition diagram.
P0' (t ) (1 c ) P0 (t ) 1 P1 (t ) c P2 (t ) (1)
'
P1 (t ) 1 P0 (t ) (2 1 ) P1 (t ) 2 P2 (t ) (2)
P2' (t ) c P0 (t ) 2 P1 (t ) ( 2 c ) P2 (t ) (3)
Initial conditions, P0 (0) 1 , and other state probabilities are zero at t = 0 (4)
Taking Laplace transformation of equations (1) to (3) and using equation (4), we obtain
2 2
r1 rl m1 r2 r2 l1 m1 m1
P0 (t ) 1 1
exp( rt
1
) exp( r2 t ) (5)
r1 ( r1 r2 ) r2 ( r1 r2 ) r1 r2
r1l2 m2 r2 l2 m2 m2
P1 (t ) exp( r1t ) exp( r2t ) (6)
r1 ( r1 r2 ) r2 ( r1 r2 ) r1 r2
r1l3 m3 r2 l3 m3 m3
P2 (t ) exp( r1t ) exp( r2 t ) (7)
r1 ( r1 r2 ) r2 ( r1 r2 ) r1 r2
Where
l1 2 1 2 c
l2 1
l3 c
m1 2 c 1 c 1 2
m2 1c 1 2 c 2
m3 2 c c 1 12
r1 , r2 (1 2 c 1 2 c ) sqrt[(1 2 c 1 2 c ) 2 4( 2 (1 c 1 ) c (1 2 1 )
2 (1 c ) c 1 )]...........(8)
Estimation of Mean Time between Failures
In this section, we have used maximum likelihood estimation approach to estimate the reliability measures such
as mean time between failures of two unit repairable systems in the presence of NCCS and LCCS failures for both series
and parallel cases.
Let the samples x1 , x2 ,........., xn ; y1 , y2 ,........., yn and w1 , w2 ,........., wn with size ‘n’ representing times between
individual, NCCS and LCCS failures which will obey exponential population respectively.
Let z11 , z12 ,........., z1n ; z 21 , z22 ,........., z2 n & z31 , z32 ,........., z3 n be ‘n’ number of times between repairs of the units with
exponential population law.
Where, xˆ , yˆ , w
ˆ , zˆ , zˆ & zˆ are the M L estimates of individual failure rate , NCCS failure rate (β), LCCS failure rate
1 2 3
(ω) and repair rates of (µ1, µ2, µc) of the system respectively.
1 1 ˆ 1 ˆ 1
Where, xˆ ; yˆ ; w ; z1 ;
x y w z1
1 1
zˆ2 ; zˆ3 ; x
x i
;y
y i
z2 z3 n n
w
w ;z z i 1i
; z2
z 2i
; z3
z 3i
1
n n n n
Volume 10, Issue 8, August 2021 191 http://adalyajournal.com/
ADALYA JOURNAL https://doi.org/10.37896/aj10.8/020 ISSN NO: 1301-2746
Series System
In this case, we have reliability function for two unit system with LCCS and NCCS failures as
RLNS (t ) P0 (t )
1
(2
r1 ) exp( r1t ) (2 r2 ) exp( r2 t ) (9)
( r1 r2 )
Where r1, r2 obtained in (8)
The mean time between failure function is
ELNS (T ) RLNS (t ).dt
0
2 / [2 (1 c ) 2 (1 c 1 ) c (1 2 1 ) c 1 ] (10)
Therefore, the M L estimate of MTBF function is given by
ˆ 2 wˆ ) / [( yp
Eˆ LNS (T ) ( yp ˆ 2 wˆ )(2( xˆ ypq
ˆ ) xˆ yp
ˆ ) zˆ (2( xˆ ypq
ˆ ) xˆ yp
ˆ zˆ )
2 1
zˆ3 (2( xˆ ypq
ˆ ) yp
ˆ 2 wˆ zˆ ) ( xˆ yp
1
ˆ ) zˆ ]..................(11)
1
Whereas xˆ , yˆ , w
ˆ , zˆ , zˆ & zˆ are the sample estimates.
1 2 3
Parallel System
In this case, we have reliability function for two unit system with LCCS and NCCS failures as
1
RLNP (t ) [(1 2 r1 ) exp(r1t ) ( 1 2 r2 ) exp( r2t )]................(12)
(r1 r2 )
Where r1 and r2 are seen in (8)
The mean time between failure function is
ELNP (T ) RLNP (t ).dt
0
( 1 2 ) / [ 2 (1 c 1 ) c (1 2 1 ) 2 (1 c ) c 1 ]..............(13)
In this case, the M L estimate of MTBF function is given by
Eˆ LNP (T ) [2( xˆ ypq ˆ 2 wˆ )] / [ zˆ (2( xˆ ypq
ˆ ) ( yp ˆ ) xˆ yp
ˆ zˆ1 ) zˆ3 (2( xˆ ypq
ˆ ) yp
ˆ 2 wˆ zˆ )
2 1
ˆ 2 wˆ )(2( xˆ ypq
( yp ˆ ) xˆ yp
ˆ ) ( xˆ yp
ˆ ) zˆ ]..............(14)
1
Simulation and Validity
In this section, the proposed estimates of the MTBF functions by M L E method do not find analytic shape of
density or distribution. Hence it is not likely to try or extend analytical verification properties of proposed M L estimates
such as ELNS (T ), E LNP (T ) of the present model. We have attempted empirical approach by Monte-Carlo simulation to
generate large samples of different sizes and mean squared error of the estimates have computed. The various sample of
sizes ‘n’ from 5 to 30 at step 5 were developed empirically with specified parameter value of failure rate and used them to
obtain MSE for various sizes of samples with simulations. The detailed analysis is tabulated and is seen in table 1 &
table 2.
For illustration purpose by fixing a range of particular value of the rate of individual (λ), NCCS failures (β),
LCCS failures (ω) and repairs (µ1, µ2, µc) for the sample size n=5(5)30 were simulated using C-Language and the sample
estimates are computed for N=10,000(20,000)90,000 and mean squared error of the estimates are evolved for
ELNS (T ), ELNP (T ) which gives reasonably small even for the small samples of size n=5. Therefore the research results
established that maximum likelihood estimation approach is satisfactory to estimate reliability indices.
Volume 10, Issue 8, August 2021 192 http://adalyajournal.com/
ADALYA JOURNAL https://doi.org/10.37896/aj10.8/020 ISSN NO: 1301-2746
Table 1. Simulation results for MTBF function in the case of series system with = 0.1, β = 0.5, 𝝎 = 0.1, µ1 = 1,
µ2 = 1.5, µc = 2, p = 0.7
N Sample ELNS (T ) Eˆ LNS (T ) M.S.E
Size
10,000 5 0.048959 0.041318 0.000448
10 0.048959 0.043380 0.000236
15 0.048959 0.044162 0.000161
20 0.048959 0.044326 0.000127
25 0.048959 0.044507 0.000104
30 0.048959 0.044595 0.000090
30,000 5 0.048959 0.041478 0.000457
10 0.048959 0.043287 0.000236
15 0.048959 0.044089 0.000165
20 0.048959 0.044368 0.000126
25 0.048959 0.044602 0.000105
30 0.048959 0.044753 0.000088
50,000 5 0.048959 0.041448 0.000458
10 0.048959 0.043368 0.000239
15 0.048959 0.044081 0.000164
20 0.048959 0.044337 0.000127
25 0.048959 0.044637 0.000103
30 0.048959 0.044718 0.000089
70,000 5 0.048959 0.041517 0.000456
10 0.048959 0.043406 0.000238
15 0.048959 0.044050 0.000163
20 0.048959 0.044427 0.000128
25 0.048959 0.044616 0.000104
30 0.048959 0.044787 0.000089
90,000 5 0.048959 0.041462 0.000459
10 0.048959 0.043366 0.000239
15 0.048959 0.044039 0.000165
20 0.048959 0.044461 0.000126
25 0.048959 0.044616 0.000104
30 0.048959 0.044747 0.000089
Volume 10, Issue 8, August 2021 193 http://adalyajournal.com/
ADALYA JOURNAL https://doi.org/10.37896/aj10.8/020 ISSN NO: 1301-2746
Table 2. Simulation results for MTBF function in case of parallel system with = 0.1, β = 0.5, 𝝎 = 0.1, µ1 = 1, µ2= 1.5,
µc = 2, p = 0.7
N Sample ELNP (T ) Eˆ LNP (T ) M.S.E
Size
10,000 5 0.107142 0.090520 0.001792
10 0.107142 0.094971 0.000973
15 0.107142 0.096789 0.000671
20 0.107142 0.096983 0.000532
25 0.107142 0.097449 0.000440
30 0.107142 0.097609 0.000384
30,000 5 0.107142 0.090907 0.001830
10 0.107142 0.094900 0.000985
15 0.107142 0.096504 0.000690
20 0.107142 0.097089 0.000528
25 0.107142 0.097711 0.000442
30 0.107142 0.097887 0.000374
50,000 5 0.107142 0.090646 0.001837
10 0.107142 0.095044 0.000981
15 0.107142 0.096547 0.000684
20 0.107142 0.097099 0.000533
25 0.107142 0.097681 0.000433
30 0.107142 0.097881 0.000375
70,000 5 0.107142 0.090801 0.001822
10 0.107142 0.094989 0.000982
15 0.107142 0.096358 0.000681
20 0.107142 0.097249 0.000536
25 0.107142 0.097656 0.000440
30 0.107142 0.097994 0.000376
90,000 5 0.107142 0.090650 0.001823
10 0.107142 0.094937 0.000991
15 0.107142 0.096398 0.000688
20 0.107142 0.097282 0.000528
25 0.107142 0.097635 0.000438
30 0.107142 0.097925 0.000376
Volume 10, Issue 8, August 2021 194 http://adalyajournal.com/
ADALYA JOURNAL https://doi.org/10.37896/aj10.8/020 ISSN NO: 1301-2746
Conclusions
We have evaluated the estimation process which could give proper estimation practice of the trustworthiness
indices with specific indication to lethal and non-lethal CCS failures. In this connection, we well thought-out a reliability
model with repair of down system. We have derived maximum likelihood estimators of the reliability indices such as
MTBF failures of the current model. The estimates of MTBF function were developed for both series and parallel systems.
The concert of the proposed estimates has been developed in requisites of MSE using simulation. The simulation outcome
suggest that M L estimate is wisely very good and give accurate estimate even for n=5. When n tending to large MSE is
zero almost in all cases of estimates.
References
1. Alexopoulos, C., & Shultes, B. C., “Estimating reliability measures for highly-dependable Markov systems, using
balanced likelihood ratios journal of Reliability”, IEEE Transactions,
50(3), 265-280, 2001.
2. Awgichew, K., & Sagar, G. Y., “Estimation of availability measures and confidence interval for two unit system with
common cause shock failures and human errors”, IOSR Journal of Mathematics, 14(3), 52 – 59, 2018.
3. Billinton, R., & Allan, R. N., “Reliability evaluation of engineering systems; concepts and techniques”, (Plenum
Press, New York), 1983.
4. Chari, A. A., Sastry, M. P. & Madhusudhana Verma, S., “Reliability analysis in the presence of common cause shock
failures”, Micro-Electronics and reliability, 31, 15-19, 1991.
5. Levitin, G., “Incorporating common-cause failures into no repairable multi state series-parallel system analysis
Journal of Reliability”, IEEE Transactions, 50(4), 380-388, 2001.
6. Reddy, Y. R., “Reliability analysis for two unit non-identical system with CCS failures, Ph. D thesis, S. K.
University”, Anantapur, 2003.
7. Sagar, G. Y., “Markovian approach to system reliability measures with common cause shock failures and human
errors”, Ph. D thesis, S. K. University, Anantapur, 2007.
8. Sagar, G. Y., Awgichew, Kifle., Melkamu, M. F., & Abdulfeta, S. M., “Simulation Study on Reliability Estimates of
a Repairable System with Lethal and Non-Lethal Common Cause Shock Failures”, Elixir Statistics, 126, 52481-
52484, 2019.
9. Sreedhar, B. R., Sagar, G. Y., Pushpanjali, K., & Reddy, Y. R., “M L Estimation of the reliability measures of a two
unit system in the presence of two kinds of CCS failures”, ARPN Journal of Engineering and Applied Sciences, 7(8),
980 – 986, 2012.
10. Verma, S. M., & Chari, A. A., “Availability and frequency of failures of a system in the presence of chance common
cause shock failures”, Micro-Electronics and reliability, 31, 265 – 269, 1991.
.
Volume 10, Issue 8, August 2021 195 http://adalyajournal.com/
