ISA Transactions 46 (2007) 127–130

www.elsevier.com/locate/isatrans

Reliability calculation under seasonally varying failure rate

Yuri A. Ermolin ∗
Moscow State University of Communication Means (MIIT), Obraztsov Street 15, Moscow, 127994, Russia

Received 28 September 2005; accepted 23 June 2006

Available online 8 February 2007

Abstract

This paper treats of the analytical solution of reliability problems in the special case when the failure rate acting on an object is a periodic
piecewise constant function of time. It is proposed that the nonstationary failure rate be substituted by a stationary one. The formula intended for
calculation of this fictitious equivalent stationary failure rate is derived. This makes possible the finding of approximate solutions of reliability
problems in an analytical form convenient for applications in the sphere of engineering and for the following analysis. A simplistic example is
used to demonstrate the proposed approach.
c 2007 Published by Elsevier Ltd on behalf of ISA.

Keywords: Reliability; Failure rate; Lifetime; Pipeline

1. Introduction the opportunity of either repairing or replacing some pipeline

network components.
The foundation of engineering techniques for an assessment In the case of nonstationary failures, attempts to obtain ex-
of a technical object’s reliability is the stationary failure flow pressions intended for engineering calculations lead either to
model [1,2]. This means that the failure rate does not vary cumbersome formulas inconvenient for practical implementa-
and remains strictly constant with time. For some real objects tion or to completely impenetrable mathematical difficulties. In
however, from observations, it is established that the hypothesis practice therefore, it is generally believed that the failure flow
of failure flow stationarity is contrary to fact. For example, acting on an object is stationary with a certain (most often, aver-
the mean number of failures in a unit of time for sewer aged over data observed) failure rate. In so doing, the reliability
network and water distribution system components tends to index assessments are, naturally, in error.
fluctuate with a periodicity of one year. It may be suggested
Below a more rigorous approach that provides a way of
that a similar situation exists also in the course of maintenance
calculating the equivalent stationary failure rate when the real
of oil, gas and other pipeline systems. Many researchers
failure rate is a periodic step function is proposed. The value of
explain this phenomenon in that the environmental condition
equivalent failure rates is calculated from a condition in which
variations produce changes in the rate of processes (corrosion,
the lifetime expectancy of an object being considered remains
underground stray currents, etc.) leading to failures. This effect
fixed under such substitution.
is most pronounced in moist subtropical regions for which the
It seems, such a setting of a problem may also be of interest
number of failures in a rainy season far exceeds those in a
for specialists working with general reliability issues.
dry season [3]. This gives grounds to consider the failure rate
to vary from season to season. But, when such is the case,
calculating reliability indexes by formulas implied from the 2. Failure rate model
stationarity hypothesis is incorrect mathematically. Meanwhile,
such calculations are necessary under decision making about Let the failure rate λ vary with time as shown in Fig. 1.
We express λ(t) in the analytical form:

λ , if (n − 1)τ ≤ t < (n − 1)τ + βτ ;


∗ Tel.: +7 495 487 6672; fax: +7 495 684 2990.
λ(t) = s (1)
E-mail address: karmapa@ipc.ru. λl , if (n − 1)τ + βτ ≤ t < nτ.

c 2007 Published by Elsevier Ltd on behalf of ISA.

0019-0578/$ - see front matter
doi:10.1016/j.isatra.2006.06.005
128 Y.A. Ermolin / ISA Transactions 46 (2007) 127–130

Notation

The following symbols are used:

A; A1 notations of some exponentials;
F unreliability function;
T lifetime expectancy; Fig. 1. Variation of failure rate with time.
T0 stationary lifetime expectancy;
Tl ; Tm ; Ts lifetime expectancy corresponding to larger,
mean and smaller failure rate, respectively;
a1 first term of progression;
f lifetime probability function;
n number of time period;
p survival function;
t running time;
q denominator of progression;
Λ integral failure rate;
β relative part of period τ ;
λ failure rate;
λ0 ; λl ; λs equivalent, larger and smaller failure rate, Fig. 2. Plot of Λ as a function of time.
respectively;
τ time period. and, hence,
Z ∞
T = e−Λ(t) dt. (5)
Here, the following notations are used: λs and λl are the 0
smaller and the larger failure rate, respectively; τ is the time In the case being considered, it is possible to write:
period; β is the relative part of the period τ during which the
λs t, if 0 ≤ t < βτ ;

failure rate λs is acting on an object (pipeline) (0 ≤ β ≤ 1);
βτ (λs − λl ) + λl t, if βτ ≤ t < τ ;


n is the number of time periods (n = 1, 2, 3 . . .); and t is the


(βτ − τ )(λs − λl ) + λs t,



running time.
Λ(t) = if τ ≤ t < (τ + βτ ); (6)
Eq. (1) is the mathematical model of nonstationary failure
2βτ (λs − λl ) + λl t,


rates used below.

(τ + βτ ) ≤ t < 2τ ;

if



... ....

3. Lifetime of object
This relationship is plotted in Fig. 2.
There are time and numerical reliability characteristics. Substituting Eq. (6) into Eq. (5) and then integrating gives:
From theory, it is apparent that all reliability time characteristics 
1 1 h i
are interrelated, i.e. if only one among them is given in an T = (1 − e−λs βτ ) + e−λs βτ 1 − e−λl τ (1−β)
λs λl
analytical form, all others, in principle, can be obtain by some
∞
mathematical transformations. In particular [2], the survival
e−nτ [λl +β(λs −λl )] .
X
× (7)
function (reliability function) p(t) is related to the failure rate
n=0
λ(t) by the equation:
 Z t  By rearranging (see “Appendix”), Eq. (7) can be put in the
closed form:
p(t) = exp − λ(t)dt (2)  
0 1 1 1
T = (1 − e−λs βτ ) − (A − e−λs βτ ) , (8)
and the lifetime expectancy T of an object (the most 1 − A λs λl
important numerical reliability characteristic), for given p(t), where the notation
is calculated from the following expression:
Z ∞ Z ∞  Z t  A = e−τ [λl +β(λs −λl )] (9)
T = p(t)dt = exp − λ(t)dt dt. (3) is introduced.
0 0 0 This result is obtained for an arbitrary τ . Usually in actual
We determine T when λ(t) varies according to Eq. (1). practice τ = 1 year. Substituting this value into Eq. (8), we get
Denote finally
Z t  
1 1 −λs β 1 −λs β
Λ(t) = λ(t)dt (4) T = (1 − e ) − (A1 − e ) , (10)
0 1 − A1 λs λl
 Y.A. Ermolin / ISA Transactions 46 (2007) 127–130 129

where now 1.0 (1/km year); τ = 1 year and β = 0.5 (all values are
hipothetical, convenient for calculations only). The difference
A1 = e−λl −β(λs −λl ) . (11)
between the calculated value of a pipe lifetime expectancy and
Thus, for given λs , λl and β, we can calculate the lifetime 1.5 years is taken as the criterion for decision making. The
expectancy T by the cumbersome, but simple expression (10). decision rule is: if the calculated value of a lifetime expectancy
is less than 1.5 years, the pipe must be replaced; otherwise there
4. Improved failure rate calculations is no need for a replacement.
First we evaluate the true value of pipe lifetime expectancy T
However, when it comes to determining time reliability char- using the data above. Calculating A1 according to formula (11)
acteristics like survival, unreliability and lifetime probability and substituting this value into expression (14), we have λ0 =
functions, the direct employment of λ(t) in the form of Eq. (1) 0.705 1/year, and, consequently, T0 = 1/λ0 = 1.418 years
leads to expressions that are unfeasible for practical calcula- that coincides, naturally, with the theoretically true value of
tions. To avoid this inconvenience, the following new approach the lifetime expectancy T . The corresponding conclusion is:
is proposed. because T0 < 1.5 years the pipe must be replaced.
Let us consider an exponential survival function correspond- It is of interest to analyze how a value of λ0 chosen
ing to the stationary failure rate: differently from that calculated by formula (14) affects the
accuracy of the determination of lifetime expectancy for the
p(t) = e−λ0 t , (12) pipe considered and, therefore, on the decision concerning
where λ0 is the constant failure rate. its replacement following from the obtained value. For this
Substituting the expression (12) in Eq. (3), for the lifetime purpose, we examine three approaches that are frequently
expectancy T0 we have: encountered in practical calculations.
Z ∞ Case A. Let us suppose that λ0 has been chosen equal to the
1
T0 = e−λ0 t dt = . (13) smaller value of the failure rate, i.e. λ0 = λs = 0.5 1/year.
0 λ 0 Then, for the lifetime expectancy Ts we get: Ts = 1/λs =
Now, we fit λ0 so that the condition T0 = T would be fulfilled. 2 years. The decision following from this estimate is: the pipe
From Eqs. (10) and (13), on rearrangement, we get: does not need replacement.
λs λl (1 − A1 ) Case B. Assume that λ0 = λl = 1.0 1/year. We have Tl =
λ0 = . (14) 1/λl = 1 year. Therefore the decision is: the pipe must be
λl − λs A1 − (λl − λs )e−λs β
replaced.
In fact, the real nonstationary failure flow is superseded Case C. Let λ0 is taken equal to the mean value between λs
formally by an improved (equivalent) stationary flow having and λl , i.e. λ0 = λm = (λs + λl )/2. Then Tm = 1/λm =
the failure rate λ0 the value of which is calculated according 1.5 year; the corresponding decision is: the pipe does not need
to Eq. (14). Under such a substitution, the lifetime expectancy replacement.
of an object remains the same. Because, for given λs , λl and β, Thus, any value of the equivalent stationary failure rate taken
λ0 is a number, for time reliability characteristics we can write different from that calculated according to the proposed new
now approximately: procedure causes the lifetime expectancy determination to be
- the survival function: p(t) ≈ e−λ0 t ; in error and may cause a faulty decision (from the three cases
- the unreliability function: F(t) ≈ 1 − e−λ0 t ; discussed above, the decision for case B only is correct).
- the lifetime probability function: f (t) ≈ λ0 e−λ0 t .
Thus, for most important reliability characteristics analytical 6. Conclusion
dependencies that are greatly convenient for practical
calculations are obtained. Classical reliability theory, in principle, provides a way
of finding all numerical and time reliability characteristics.
5. Example However, in the case that the failure rate acting on an object
is nonstationary, the fulfilment of prescribed formal procedures
The method developed in this paper may be used to solve brings results, which, as a rule, contain integrals involving
many practical reliability problems. One of these is considered expressions of very complicated functions. The calculation
below in an extremely simplified but realistic statement. of such integrals is possible only by means of numerical
Let us assume that a question concerning replacement of calculations, which is not always convenient. The new
some sewer pipe 1 km in length by a new pipe during the improved approach proposed in this paper makes it possible to
next 1.5 years is discussed. The decision is made on the basis circumvent these difficulties by using the concept of equivalent
of historical failure data obtained and processed over a long stationary failure rate. In practical engineering applications
preceding period of time. It has been established, for example, approximate analytical solutions, which are of simpler form
that the failure rate acting on the pipe varies from season and convenient for analysis, are frequently acceptable. Such an
to season and takes the form of the time-dependent function approach allows one to find the solution of reliability problems
shown in Fig. 1 with parameters: λs = 0.5 (1/km year); λl = appearing, for example, in exploring rehabilitation needs and
130 Y.A. Ermolin / ISA Transactions 46 (2007) 127–130

strategies for water distribution systems [4], in estimation of Consequently

raw sewage discharge resulting from sewer network failures [5], ∞
e−τ [λl +β(λs −λl )]
e−nτ [λl +β(λs −λl )] =
X
and so forth, in a convenient analytical form. .
n=1
1 − e−τ [λl +β(λs −λl )]
Appendix
Thus
We present the sum involved in Eq. (7) as: ∞
1
e−nτ [λl +β(λs −λl )] =
X
.
∞ ∞ n=0
1 − e−τ [λl +β(λs −λl )]
e−nτ [λl +β(λs −λl )] = 1 + e−nτ [λl +β(λs −λl )] .
X X
(A.1)
n=0 n=1
Substitution of this expression into Eq. (7) leads to Eq. (8).

Because (λs − λl ) < 0 and 0 < β < 1, the index of the References
exponential function is always negative and, consequently, the
[1] Gnedenko BW, Beliaev YuK, Soloviov AD. Mathematical methods in
second term in Eq. (A.1) is none other than the sum of terms of reliability theory. Moscow: Nauka; 1965 [in Russian].
an infinitely decreasing geometric progression. It is well known [2] Mays LW, et al., editors. Reliability analysis of water distribution systems.
that this sum S is given by the expression [6]: New York: ASCE; 1989.
a1 [3] Ckay L. Study of water distribution networks reliability and ways to its
S= , upgrading in Vietnam. Ph.D. thesis. Moscow: Moscow State University of
1−q Civil Engineering; 2002 [in Russian].
[4] Herz RK. Exploring rehabilitation needs and strategies for water
where a1 is the first term, and q is the denominator of the
distribution networks. J Water SRT - Aqua 1998;47(6):275–83.
progression. [5] Ermolin YA. Estimation of raw sewage discharge resulting from sewer
Here network failures. Urban Water 2001;3(4):271–6.
[6] Korn GA, Korn TM. Mathematical handbook for scientists and engineers.
a1 = q = e−τ [λl +β(λs −λl )] . New York, Toronto, London: McGraw-Hill Book Company Inc.; 1961.