ISSN 1330-3651 (Print), ISSN 1848-6339 (Online) https://doi.org/10.

17559/TV-20171220201513
Preliminary communication

A Stochastic Model for Achieving Required Level of Availability Based on the Repair Rate
Analysis

Nataša KONTREC, Milena PETROVIĆ, Stefan PANIĆ, Hranislav MILOŠEVIĆ

Abstract: The aim of this paper is to set framework for the repair rate in order to achieve required availability of a repairable system with alternating renewal process where
the failure rate is Weibull distributed. In order to accomplish this goal, this paper proposes a novel model for a complete stochastic analysis of the repair rate. The resulting
equation for probability density function of the repair rate enables precise modelling of the repair rate process for related values of availability and failure rate. The obtained
results can be used for dynamic forecasting of system’s characteristics, planning of maintenance activities, inventory and servicing capacities.

Keywords: alternating renewal process; availability; repair rate; stochastic analysis

1 INTRODUCTION As stated previously E(U) is MTTF, while E(D) is MTTR,

so the expression for availability can be reduced to the well
This paper considers a repairable system with failure- known equation:
repair cycles modelled by renewal processes. Such system
alternates from operational to failure state in a period of MTTF
time [1]. In the moment of failure, the repairs or A= (3)
MTTF + MTTR
replacements of parts are conducted, after which this
system returns to its operational state i.e. performs as brand Many papers dealing with the system’s availability
new. If we denote the operational state as 1, then we can assessments exist at present time. In papers [9, 10], the
present the time intervals where the system is in the author determines the steady state availability and trust
aforesaid condition as U = (U1, U2, U3, ...). Similarly, if we intervals in the case where repair time has a lognormal
denote the condition where the system is in the failure state distribution. The steady state availability and methods to
with 0 then we can present the time intervals for system determine this parameter for repairable systems have been
repairs with D = (D1, D2, D3, …). Moreover, let us assume examined in papers [11-13]. Significant contributions
that T = ((U1, D1), (U2, D2),…) is a sequence of concerning determining the availability in repairable
interdependent identically distributed intervals. Although systems which function under maintenance contracts were
these intervals are independent, Dn can be expressed with given in papers [14-18]. They acknowledged that the
the use of Un(i.e. for n ∈ N+. Dn can be presented in system’s availability primarily depends on the system’s
function of Un), and vice versa. Therein E(U) represents the reliability, MTTF, number of spare parts in inventory and
mean time to failure (MTTF) while E(D) represents the mean time required to return the system to operational
mean time required for the system to return to the state. A similar problem was further examined in paper
operational state i.e. the mean time to repair (MTTR). [19] concluding that the system’s repair time has the
Several important applications of the alternating renewal highest effect on system’s availability. Due to that reason,
process are provided in papers [2] and [3]. we have decided to examine the system’s repair time as a
The alternating renewal process is often used in order stochastic process, with the aim to determine the repair rate
to determine the availability of a particular system, as the without having the system’s availability level falling below
most common measure of efficiency of repairable systems the set limits.
[4, 5]. Numerous factors affect the availability, such as
system’s reliability, spare parts’ availability, system’s 2 MATHEMATICAL MODEL FOR REPAIR RATE
maintenance etc. The availability is defined as a probability ANALYSIS
that the system will perform its function in a period of time
[6]. The limiting availability or steady state availability is The model presented in this paper is based on several
most commonly used to determine the long-run assumptions. Firstly, the observed repairable system can be
performance of a system [7]. modelled with the alternating renewal process. Then, we
are assuming that the system behaves like brand new, after
A = lim A(t ) (1) the repairs had been made. Moreover, the failure rate has
t →∞
the Weibull distribution. In papers [20, 21] the failure rate
If the key renewal theorem is applied, as presented in was modelled by the Rayleigh distribution but according to
paper [8], the previous formula can be reduced as follows: [22-25] this parameter can also be modelled by using the
Weibull distribution. Actually, according to [26], the
failure rate of mechanical and electrical components has a
E (U )
A= . (2) bath-tub shape. It was shown that the Weibull distribution
E (U ) + E ( D ) is very flexible in modelling various types of lifetime
distributions so it can be used to model any of the three
parts in a bath-tub curve (early "infant mortality" failure,

Tehnički vjesnik 26, 4(2019), 1171-1175 1171

 Nataša KONTREC et al.: A Stochastic Model for Achieving Required Level of Availability Based on the Repair Rate Analysis

constant (random) failure and increasing ("wear out") 2


failure. Γ 1 +

According to the aforementioned, if the system’s x=  β . (10)
failure time is Weibull distributed, the MTTF can be  1
Γ 1 +  λ
determined with the following Eq. (4):  β 

∞ Determining the system’s failure rate is a complex

MTTF
= ∫=
tpWeibull (t )dt process and due to its stochastic nature the random variable
0 x can be considered as random variable that changes at
β
(4)
∞ β −1
βt  t   1 significantly slower rate than random variable t described
=∫t ω β exp  − ω  dt =
Γ  1 +  ω.
 β 
with the Weibull model. As is known from the theory of
0
stochastic processes [28] in the case where x ≈ t 2 and
random variable t is described with the Weibull model,
Let denote λ as a measure of failure rate as presented
then the slow changes of the random variable x can be
in [27]:
presented as a stochastic process with exponential
distribution as in the Eq. (11)
1 1
=λ = =
β
MTTF ∞
βtβ −1
 t   x 
∫ t ω β exp  − ω  dt exp  − 
0
(5) px (x)
=  x0  , x >=
0, x0 E (x). (11)
1 x0
= .
 1
Γ 1 + ω
 β  The probability density function (PDF) of failure rate
can now be presented with the following transformation:
In the Eqs. (4) and (5), ω and β are parameters of the
Weibull distribution −β denotes shape parameter while ω   2 
denotes characteristic life parameter. Since the system Γ 1 + β  
p ( λ ) = px     dx , (12)
repair rate, denoted as µ, is equal to the reciprocal value of   1   dλ
the mean time to repair (MTTR), the Eq. (3) can be Γ 1 + β  
presented as:   

µ dx
A= . (6) where can be determined as in (13)
µ +λ dλ

Now, knowing that the second moment of Weibull  2

distribution is: Γ 1 + 
dx  β
 .
= (13)
 2 dλ  1 2
E t2( ) =Γ 1 +  ω =x
 β
(7) Γ 1 +  λ
 β

we can determine the distribution parameteras: Now, by replacing the Eq. (12) and (13) into (11), we
get the following equation failure rate PDF:
x
ω= . (8)   2 
 2  Γ 1 +  
Γ 1 +   2  β 
 β Γ 1 +  exp  − 
 β   1 
 Γ 1 +  λ x0 
By replacing ω in the Eq. (5) the failure rate transforms   β .
p (λ ) = (14)
into:  1 2
Γ 1 +  λ x0
 β
2 
Γ 1 +
β 
λ=  . (9) Based on the Eq. (6), the failure rate is λ =
1− A
µ with
 1 A
Γ 1 +  x
 β dλ 1 − A
the Jacobian transformation = while the repair
dµ A
Using the Eq. (9) we can calculate the value of x as in
the Eq. (10) Aλ
rate can be expressed as µ = with PDF:
1− A

1172 Technical Gazette 26, 4(2019), 1171-1175

 Nataša KONTREC et al.: A Stochastic Model for Achieving Required Level of Availability Based on the Repair Rate Analysis

 A  dλ  2
p ( µ ) = pλ  . (15) Γ 1 +  A
  β  1
 1 − A  dµ µ= . (20)
 1  2 
Γ 1 +  ( A − 1) x0 ln  (1 − A ) y 
According to the previous equations we can now  β  A2 
determine the PDF repair rate as in (16)  

p(µ ) = 3 NUMERICAL RESULTS

 2   2  In order to verify the mathematical model from the

Γ 1 + A  Γ 1 + A  (16)
 β   β  previous section we have used data from the case study
exp  − .
 1   1  presented in paper [29]. The MTTF data for the following
Γ 1 +  (1 − A ) µ 2 x0  Γ 1 + β  (1 − A ) µ x0  subsystems for railway signalling are provided in this
 β    
study:
- Interlocking: As input, receives data from different
At the same time, the Eq. (16) presents one of the systems and calculates output which is then used to
major contributions stated in this paper. It provides a improve traffic safety
precise mathematical characterization of a stochastic - Track circuit: Used to locate trains
process that describes the repair time. The PDF expression - Balise group: Sends data on speed limits, driving
for the repair rate enables precise modelling of the repair limitations etc. from tracks to signalization system
rate process for related values of availability. This can be - Level crossing: Coordinates railway traffic
used for dynamic forecasting of system’s characteristics - Signalling board: Provides timely information
through simulation of the repair rate process by generating regarding tunnels, bridges, speed limits etc.
samples.
In order to further analyze the repair rate, the
cumulative distribution function (CDF) is provided:

F (µ ) =
 2   2 
Γ 1 +  A µ  Γ 1 +  A  (17)
 β 1  β 
 1
∫ µ 2  −
exp
 1
 dµ .

Γ 1 +  (1 − A ) x0 0  Γ  1 + β  (1 − A ) µ x0 
 β    

By solving the integral in the previous equation, the

expression for the CDF function can be presented as:

   2 
 2  Γ 1 +  A 
A 1 − exp  −  β    . (18)
(µ )
F= 2   
(1 − A)   1 
 Γ 1 +  (1 − A ) µ x0   Figure 1 PDF of failure rate for Balise group
  β
  

In order to obtain more precise data on the repair rate

we can now apply the inverse sampling method. If we state
that y = F ( µ ) where y is uniformly distributed in interval
[0,1] then the inverse CDF is F −1 ( µ ) = y −1 . Having this in
mind, we can transform the previous equation into:

 2
 (1 − A )2  Γ 1 +  A
β
ln 
 A2
(1 − y )  =  , (19)
 1
  Γ 1 +  ( A − 1) µ x0
 β

and determine the repair rate samples as in (20)

Figure 2 PDF of failure rate for Interlocking

It was determined that the MTTF for subsystem

Interlocking is 2.8581 years, for Balise group 4.767, for

Tehnički vjesnik 26, 4(2019), 1171-1175 1173

 Nataša KONTREC et al.: A Stochastic Model for Achieving Required Level of Availability Based on the Repair Rate Analysis

Level crossing 2.286, for Signal 2,4663 and for Track availability. The system’s availability in this specific
circuit 2.004. example is 80%, 90% and 98% in proportion to time the
system is in operational condition. It is possible to take into
consideration other values of availability depending on the
observed system and previously set demands. Likewise,
Figs. 2, 3, 4 and 5 represent the PDF of failure rates for
other sub-systems: Interlocking, Level crossing, Signal and
Track circuit.
Based on the previously presented figures we can
determine the probability that the repair rate will take one
of the offered values. That information, together with the
CDF inverse sampling, can be useful for planning the
system’s maintenance activities, number of service
stations, spare parts and manpower required for
maintenance.

5 CONCLUSION

Related researches of repairable systems’ maintenance

Figure 3 PDF of Level crossing processes showed that both reliability and repair rate have
significant impact on availability of such systems. In this
paper we studied a system which can be modelled with the
alternating renewal process. We assumed that the failure
rate is Weibull distributed and that the MTTF is a
predetermined value. Also, after repairs, the system
returned to its original state and performed as new. In
accordance with these assumptions, a novel model for
determining the characteristics of the repair rate was
formulated. We determined the expression for the repair
rate‘s PDF and CDF which represents a precise
mathematical characterization of the repair rate as a
stochastic process. Using this expression can result in exact
repair rate sample values for corresponding values of
availability. In this way, by simulating the repair rate
process through generating its samples, we can predict the
system‘s dynamic characteristics. In Numerical result
section, the previously described model was applied to a
specific example i.e. five sub-systems of Railway
Figure 4 PDF of Signal Signalling System and graphical presentation of the PDF
for each sub-system was provided. This model can be
applied in the same manner to other repairable systems
which can be modelled with the alternating renewal
process. In addition, other values can be selected for
availability. The obtained results can be further utilized in
planning the maintenance activities, inventory, service
systems and number of required employees, in the process
of system maintenance.

Acknowledgments

This work was funded by the framework of

Competitiveness Enhancement Program of the National
Research Tomsk Polytechnic University, Russia and
supported in part by the Serbian Ministry of Education,
Science and Technological Development, Projects #TR
35030.
Figure 5 PDF of failure Track circuit

Based on the previous model, the PDF function 6 REFERENCES

depending on the required availability was presented for
each sub-system. [1] Wolstenholme L. C. (1999). Reliability Modeling. A
Stochastic Approach, CRC Press, USA.
Fig. 1 represents the PDF for the sub-system of Balise
[2] Barlow, R. E. & Proschan, F. (1965). Mathematical Theory
group for randomly selected values of sub-system of Reliability. New York, NY, USA: Wiley

1174 Technical Gazette 26, 4(2019), 1171-1175

 Nataša KONTREC et al.: A Stochastic Model for Achieving Required Level of Availability Based on the Repair Rate Analysis

[3] Nakagawa, T. (2005) Maintenance Theory of Reliability. [20] Kontrec, N., Milovanović, G., Panić, S., & Milošević, H.
London, U. K.: Springer-Verlag (2015). A Reliability-Based Approach to Nonrepairable
[4] Pham‐Gia, T. & Trukkan, N. (1999), System availability in Spare Part Forecasting in Aircraft Maintenance
a gamma alternating renewal process. Naval Research System, Mathematical Problems in Engineering, 2015, 1-7.
Logistics, 46, 822-844. https://doi.org/10.1002/(SICI)1520- https://doi.org/10.1155/2015/731437
6750(199910)46:7<822::AID-NAV5>3.0.CO;2-D [21] Kontrec, N., Panić S., Petrović, M., & Milošević, H. (2018).
[5] Yao, K. & Gao, J. (2015). Uncertain random alternating A stochastic model for estimation of repair rate for system
renewal process with application to interval availability. operating under performance based logistics. Eksploatacja I
IEEE Trans. Fuzzy Syst., 23(5), 1333-1342. Niezawodnosc – Maintenance and Reliability, 20(1), 68-72.
https://doi.org/10.1109/TFUZZ.2014.2360551 https://doi.org/10.17531/ein.2018.1.9
[6] Barlow, R. E. & Proschan, F. (1975). Statistical theory of [22] Kontrec, N. & Panić, S. (2017). Spare Parts Forecasting
reliability and life testing. Probability Models. New York: Based on Reliability, System Reliability, Constantin
Holt, Rinehart and Winston. Volosencu, IntechOpen.
[7] Gamiz, M. L., Kulasekera, K. B., Limnios, N. & Lindqvist https://doi.org/10.5772/intechopen.69608
B.H. (2011). Applied Nonparametric Statistics in Reliability. [23] Mudholkar, G. S. & Srivastava, D. K. (1993). Exponentiated
Springer-Verlag London. Weibull family for analyzing bathtub failure rate data. IEEE
https://doi.org/10.1007/978-0-85729-118-9 Transactions on Reliability, 42, 299-302.
[8] Ross, S. M. (1992). Applied probability models with https://doi.org/10.1109/24.229504
optimization applications. New York: Courier Dover [24] Xie, M., Tang, Y., & Goh, T. N. (2002). A modified Weibull
Publications extension with bathtub-shaped failure rate function.
[9] Ananda, M. M. A. (2003). Confidence intervals for steady Reliability Engineering & System Safety, 76(3), 279-285.
state availability of a system with exponential operating time https://doi.org/10.1016/S0951-8320(02)00022-4
and lognormal repair time. Applied Mathematics and [25] Carrasco, J. M. F., Ortega, E. M. M., & Cordeiro, G. M.
Computation, 137(2-3), 499-509. (2008). A generalized modified Weibull distribution for
https://doi.org/10.1016/S0096-3003(02)00155-8 lifetime modeling. Computational Statistics & Data
[10] Ananda, M. M. A. & Gamage, J. (2004). On steady state Analysis, 53(2), 450-462.
availability of a system with lognormal repair time. Applied https://doi.org/10.1016/j.csda.2008.08.023
Mathematics and Computation, 150(2), 409-416. [26] Xie, M. & Lai, C. D. (1996). Reliability analysis using an
https://doi.org/10.1016/S0096-3003(03)00281-9. additive Weibull model with bathtub-shaped failure rate
[11] Claasen, S. J., Joubert, J. W., & Yadavalli, V. S. S. (2004). function. Reliability Engineering and System Safety 52, 87-
Interval estimation of the availability of a two unit standby 93. https://doi.org/10.1016/0951-8320(95)00149-2
system with non instantaneous switch over and "dead time". [27] Pham, H. (2006). System Reliability Concepts. In: System
Pakistan Journal of Statistics, 20(1), 115-122. Software Reliability. Springer Series in Reliability
[12] Hwan Cha, J., Sangyeol, L., & Jongwoo, J. (2006). Engineering. Springer, London.
Sequential confidence interval estimation for system https://doi.org/10.1007/1-84628-295-0_2
availability. Quality and Reliability Engineering [28] Leemis, L. & McQueston, J. (2008) Univariate distribution
International, 22, 165-176. https://doi.org/10.1002/qre.693 relationships. The American Statistician, 62, 45-53.
[13] Ke, J. C. & Chu, Y. K. (2007). Nonparametric analysis on https://doi.org/10.1198/000313008X270448
system availability: confidence bound and power function. [29] Morant, A., Gustafson, A., Söderholm, P., Larsson-Kråik,
Journal of Mathematics and Statistics, 3(4), 181-187. P., & Kumar, U. (2016). Safety and availability evaluation
[14] Kang, K., Doerr, K. H., & Sanchez, S. M. (2006). A Design of railway operation based on the state of signalling systems.
of Experiments Approach to Readiness Risk Analysis. Proceedings of the Institution of Mechanical Engineers, Part
Proceedings of the 38th conference on Winter F: Journal of Rail and Rapid Transit, 231(2), 226-238.
simulation (WSC '06), L. Felipe Perrone, Barry G. Lawson, https://doi.org/10.1177/0954409715624466
Jason Liu, and Frederick P. Wieland (Eds.). Winter
Simulation Conference, 1332-1339.
https://doi.org/10.1109/WSC.2006.323232 Contact information:
[15] Kim, S. H., Cohen, M. A., & Netessine, S. (2007)
Performance contracting in after-sales service supply chains. Nataša KONTREC, Assistant Professor
Management Science, 53(12), 1843-58. (Corresponding author)
https://doi.org/10.1287/mnsc.1070.0741 Faculty of Sciences and Mathematics, University of Priština,
[16] Kim, S. H., Cohen, M. A., & Netessine, S. (2007). Lole Ribara 29, 38220, Kosovska Mitrovica, Serbia
natasa.kontrec@pr.ac.rs
Reliability or inventory? Contracting strategies for after-
sales product support. Working Paper. OPIM Department, Milena PETROVIĆ, Assistant Professor
The Wharton School. Faculty of Sciences and Mathematics, University of Priština,
[17] Nowicki, D., Kumar, U. D., Steudel, H. J., & Verma, D. Lole Ribara 29, 38220, Kosovska Mitrovica, Serbia
(2008). Spares provisioning under performance-based milena.petrovic@pr.ac.rs
logistics contract: profit-centric approach. Journal of the
Operational Research Society, 59(3), 342-52. Stefan PANIĆ, Associate Professor
https://doi.org/10.1057/palgrave.jors.2602327 (1) National Research Tomsk Polytechnic University,
Sovetskaya 84/3, 634050 Tomsk, Russian Federation
[18] Öner, K. B., Kiesmüller, G. P. & van Houtum, G. J. (2010).
(2) Faculty of Sciences and Mathematics, University of Priština,
Optimization of component reliability in the design phase of Lole Ribara 29, 38220, Kosovska Mitrovica, Serbia
capital goods. European Journal of Operational Research, stefanpnc@tpu.ru
205(3), 615-24. https://doi.org/10.1016/j.ejor.2010.01.030
[19] Mirzahosseinian, H. & Piplani, R. (2011). A study of Hranislav MILOŠEVIĆ, Professor
repairable parts inventory system operating under Faculty of Sciences and Mathematics, University of Priština,
performance-based contract. European Journal of Lole Ribara 29, 38220, Kosovska Mitrovica, Serbia
Operational Research. 214(2), 256-61. hranislav.milosevic@pr.ac.rs
https://doi.org/10.1016/j.ejor.2011.04.035

Tehnički vjesnik 26, 4(2019), 1171-1175 1175