tom XLII(2012), nr 3-4, 15–29

Janusz Buchta∗
Andrzej Oziemski
Lodz University of Technology
Institute of Electrical Power Engineering
Łódź

Probabilistic methods in the reliability assessment

of power units

The paper presents results of reliability analysis made for a lignite-fired 370 MW rated power
units installed in the Bełchatów Power Station. The concept of a standardized power unit and
the method of a histogram with a set number of observations in each class were applied. The
study includes analysis of probability distributions of operation times and repair times for the
main power unit components. Empirical probability distribution functions have been identified
and their parameters estimated in the study. The final forecast includes an estimation of such
reliability measures like expected operation time, expected failure rate, average repair time and
expected annual failure duration.

1 Introduction
Contemporary industry based on complex technological processes sets high re-
quirements concerning an electrical power quality and continuity of supply. The
complexity of the power system as well as power concentration in power stations
make difficult the realization of these requirements. Rated power of a single units
under construction still grows and this causes the rise of the emergency of peak
power shutdown and limitations in power supply in consequence.
Electric power blackouts experienced in United States, Great Britain and Italy
reminded the importance of power system reliability. These incidents have shown
that the occurrence of a massive outage is feasible and can happen in each coun-
try. The superior target of the power system is to guarantee the continuity of
supply for all consumers as well as required power quality at lowest generation
costs.
∗
E-mail: janusz.buchta@p.lodz.pl
16 J. Buchta and A. Oziemski

2 Mathematical basis of the reliability estimation of

power units by the method of a histogram

The operation practice of power system demonstrates that failure frequency of

power units is many times higher than other power system components (i.e. over-
head lines, transformers, switchgear, protections etc.) [1,11]. The characteristic
feature of power unit reliability, directly resulting from the redundancy of an aux-
iliary systems is the possibility to appear failures of different kind, i.e. power unit
continues operation with its rates, power units operates with power limitation or
power unit must be shut-down. This means that power unit is a multi-stage ob-
ject in sense of its reliability, on the contrary to the two-stage installations staying
in ability or inability to operate, which refers to many power transmission and
distribution devices.
Estimation of reliability in power generation process requires the knowledge
of reliability rates both power unit components and entire power plant. One
should point out that power unit components have various durability. Some of
them could be in the lifetime period with decreasing or stable failure frequency
when others have achieved ageing period characterizing with growth of failure
frequency. Complicated power installations are processed by intensive fatigue or
wear and tear (metal erosion, moisture, temperature, pressure) which affect on
installation elements in a different degree.
The Bełchatów Power Station, which is the subject of the study, is the largest
lignite fueled power station in Europe with capacity of 5298 MW. The power
station consists of twelve 370 MW and one 858 MW power unit and has around
20reliability analysis is focused on twelve 370 MW power units which techno-
logical structure is typical for thermal power station and include steam boiler,
turbine, generator, auxiliaries etc. All units are homogenous in several respects
such as constructional uniformity and similar conditions in which units operate.
Since the first of dozen 370 MW units was commissioned in the Bełchatów Power
Station (Poland) in 1982, a systematic research on power units reliability was ini-
tiated by the Institute of Electrical Power Engineering at Technical University of
Łódź [5–7]. The principal target of analyses prepared yearly has been estimation
of actual reliability measures of main generating devices of power units and their
most defective elements, especially heating surfaces in steam boiler BB-1150.
Statistical data files of the successive years of the power station operation have
been systematically complemented and verified. The verification of statistical
data consist in elimination of events which are not of random origin (i.e. ac-
tively influenced by operation and maintenance staff) and these, which were not
Probabilistic methods in the reliability assessment. . . 17

qualified as break downs only because there was enough ready-reserve power dur-
ing failure. Using reliability model (Fig. 1) and computer database worked out
for the 370 MW power unit, the most defective elements of the boiler, turbine,
generator and auxiliary systems have been identified [3,4]. For selected elements,
empirical probability density functions of operation times, repair times have been
determined with use of the histogram with a set number of observations in each
class.

Figure 1. Schematic diagram of the three-step decomposition of the power unit for reliability
model.

This histogram is prepared according to the following steps [12]:

• sorting numbers in a data set of n observations in an ascending order t1 ,

t2 , t3 , . . . tn ;
• calculating number of classes r by the formula

r = 2 ln(n) (1)

and rounding down r to the nearest integer;

18 J. Buchta and A. Oziemski

• calculating number of observations m in each class by the formula

n
m= (2)
r

and rounding down m to the nearest integer;

• assigning all observations to each class in such a way that equal numbers
from a data set are included in the same class; therefore the number of
observations in the i-th class, ni , can vary from an assumed value m;
• determining the lower and upper bound of each class; for example the upper
bound of the 4th class (lower bound of the 5th class) is an arithmetic mean
of the greatest number included in the 4th class and the lowest number
included in the 5th class;
• calculating the i-th class interval width, ∆ti as the difference of its upper
and lower bound; determining the i-th class midvalue, t̄i , as an arithmetic
mean of its lower and upper bounds;
• calculating the value of an empirical probability density function by the
formula
ni
f x (t̄i ) = . (3)
n∆ti

The histogram is completed by drawing a bar for each class. The main advantage
of this histogram is the possibility to apply for the amount of classes r ≥ 7 and
less numerous data sets. If take into account the minimum required amount of
observations in a single class (ni ≥ 5), the minimum statistical sample (n ≥ 35)
is easy to achieve in practice.
Calculation unit was implemented in computer database to identify prob-
abilistic models for empirical distributions of operational times and shutdown
times of standardized 370 MW power unit. Considering times of failures recorded
in a computer database, times to failure and times of failure duration have been
calculated. An empirical probability density function f x (t) is compared with the
shape of the density function of different theoretical distributions (exponential,
Weibull, normal, log-normal), and subjectively best distribution is selected as the
one representing the random variable under investigation [8,10]. The hypothesis
thus constructed is verified by means of (Pearson and Kolmogorov) statistical
tests of goodness of fit, and only on this basis is the decision made to accept
or reject it. All calculations are performed for the standard significance level
α = 0.05. Probability density functions f (t), cumulative distribution functions
F (t) and means E(T ) for considered distributions are presented in Tab. 1. Esti-
mators of distribution parameters are presented respectively in Tab. 2 [9].
Probabilistic methods in the reliability assessment. . . 19

Table 1. Probability density functions f (t), cumulative distribution functions F (t) and means
E(T ) for considered distributions.

Distribution f (t) F (t) E(T )

Exponential λ exp(−λt) 1 − exp(−λt) 1
λ
λ>0
Weibull h i h i
b t b−1 t b t b 1





a>0 a a exp − a 1 − exp − a aΓ b + 1
b>0
Normal h 2
i
√1 exp − (t−m) t−m


m≥0 σ 2π 2σ 2 0.5 + Φ σ m
σ>0
Log-normal h i   h i
t−m)2 σ2
m>0
lg
√e
tσ 2π
exp − (lg 2σ 2 0.5 + Φ lg t−m
σ exp m
lg e + 2(lg e)2
σ>0
Explanations: Γ – gamma function, Φ – Laplace’s function, λ – reciprocal of the scale
parameter, σ – shape parameter, t, T – operation time or repair time as a random variable
continuous and discrete respectively, a – scale parameter, b – shape parameter.

3 Histograms of operation times and repair times for

the 370 MW power unit, its main components and
selected elements of the BB-1150 steam boiler
The idea of a standardized unit has been introduced into the reliability study of
the 370 MW units. All units are homogenous in several respects such as construc-
tional uniformity and similar conditions in which units operate. The standarized
unit is the unit in its useful life with stabilized failure rate that substitutes a dozen
of power station units. The assumption that a standarized unit is in its useful life
requires to exclude a period of an early life from an operation time of respective
power units. The concept of a standardized unit allows to obtain an appropri-
ately numerous population of failures not only for a power unit but also for its
main generating devices (boiler, turbine, generator) and boiler elements. While
establishing the population, the cases of incidental failures that occurred in the
initial period of the power plant operation were neglected. It particularly refers to
units 1 and 2 during the first three years of the power plant operation. The cases
of failures that occurred in first year of operation were also skipped for remaining
units. One can state that a higher failure frequency of a 370 MW power units in
20 J. Buchta and A. Oziemski

Table 2. Estimators of distribution parameters for considered distributions.

Distribution Parameter Estimator

n
Exponential λ λ∗ = n
P
ti
i=1
n
1 ∗
θ = ab tbi
P
θ∗ = n
Weibull i=1

nθ ∗
b b∗ = n
∗
n
tbi ln ti −θ ∗
P P
ln ti
i=1 i=1

n
1 P
m m∗ = n ti
Normal i=1
s
n
(ti −m∗ )2
P
i=1
σ σ =
∗
n−1
n
1
P
m m∗ = n lg ti
Log-normal i=1
s
n
(lg ti −m∗ )2
P
i=1
σ σ =
∗
n−1

the initial period of their operation was caused mainly by a design, construction
and assembly defects. All these factors are distinctive for an adaptation of new
generation power units. Therefore this period was excluded while population of
failures for a standardized 370 MW unit was selected.
An investigated probability distribution functions were identified as Weibull
distributions with parameter b < 1 (Fig. 2). In emergency states, strong depen-
dence of the time of damage liquidation on the cause of its occurrence was found.
Average times of shut down, in the case of permanent defects of the installations
clearly differ from the times of shut down caused by the incorrect operation of
the automatic control systems and safety devices, the operational staff mistakes
etc. So, the values of times are as follows: 41.3 h and 1.75 h for the boiler; 59.7 h
and 1.27 h for the turbine; 78.9 h and 1.48 h for the generator; 40.1 h and 1.95 h
for feed water pump, respectively. Thus, it should be concluded that the times of
long-term failures and short-term defects belong to two statistically different pop-
ulations and that it is advisable to study their distributions separately. Generally
speaking, the distributions of failure duration times are log-normal distributions
(Fig. 3).
Probabilistic methods in the reliability assessment. . . 21

Figure 2. Probability density functions of times between failures identified as Weibull distri-
butions: a) boiler BB-1150, b) turbine 18K370, c) generator GTHW-370, d) power
unit, e) evaporator, f) P1B convection superheater. Explanations: a, b – estimated
parameters of Weibull distribution, χ2 – the value of Pearson’s statistics, χ2α – the
critical value of Pearson’s statistics for the significance level α = 0.05, λ – the value of
Kolmogorov statistics, λα – the critical value of Kolmogorov statistics for the signifi-
cance level equal to 0.05, i – the class number, n – the sample size, r – the number of
classes, suma – the total value of all the observations in the sample, ni – the number
of observations in the i-th class, fi – the value of empirical probability density function
for the i-th class.
22 J. Buchta and A. Oziemski

Figure 3. Probability density functions of failures times identified as log-normal distributions:

a) generator GTHW-370 (times of long-term failures), b) generator GTHW-370 (times
of short-term failures).

Figure 4. Probability density functions of failures times: a) boiler BB-1150 – distribution

unidentified, b) evaporator – distribution unidentified, c) convection superheater P1B
– distribution unidentified, d) economiser – distribution unidentified.
Probabilistic methods in the reliability assessment. . . 23

Probability density functions of failure times for the boiler (Fig. 4a) and its
heat transfer surfaces (Figs. 4b–4d) have complicated shapes. An interesting
feature of these distributions is an occurrence of two time intervals with high
probability level of failure appearance with respective time duration. The first
interval refers to failures with duration of 20–30 h, the second with duration of
40–60 h. The belongings of the failure to one of these groups is connected with
the scope of leakages that steam boilers are affected. One can hypothesize that
failures of the boiler with time duration out of these intervals have no relation to
damages of heat transfer surfaces or that are composed by the superposition of
failures of a few elements.
An empirical distribution functions of operation and failure times for each of
dozen power units (Figs. 5 and 6) and steam boilers (Fig. 7) have been also
examined.

Figure 5. Probability density functions of times between failures identified as Weibull distribu-
tions: a) unit 1, b) unit 4, c) unit 8, d) unit 12.
24 J. Buchta and A. Oziemski

Figure 6. Probability density functions of short-term failures times identified as log-normal dis-
tributions: a) unit 1, b) unit 4, c) unit 8, d) unit 12.

Figure 7. Probability density functions of times between failures identified as Weibull distribu-
tions: a) boiler 1, b) boiler 12.
Probabilistic methods in the reliability assessment. . . 25

4 An assesment of reliability indicies of the 370 MW

power units and their main components
By means of estimation, parameters of identified probability distribution func-
tions of operation and repair times have been calculated for investigated devices
of 370 MW power units. Such reliability indices as expected failure rate, ex-
pected mean time of the shutdown, expected annual time of shutdowns have
been estimated and expected failure-free time. Table 3 presents expected values
of basic reliability indices for a standardized 370 MW power unit and its main
components. Table 4 contains the results of reliability forecasts for shutdowns
of important elements of the BB-1150 boiler, i.e. for an economizer, evaporator,
inner suspended pipes P1A, convection superheater P1B, outlet superheater PIV,
reheaters M1 and M2 and all superheaters summed up. Code symbols used for
elements of the steam boiler BB-1150 remain in accordance with manufacturer
specification and are applied in bibliography (for example [13]). Table 5 consist
of expected values of reliability indices for respective power units of the power
station whereas Tab. 6 for corresponding steam boilers.

Table 3. Reliability indices for main power plant components.

Failure Expected Mean time of Total time of Mean time

location failure rate a shutdown shutdowns between failures
[1/a] [h] [h/a] [h]
B 3.79 37.83 143.50 1626
T 0.83 13.31 11.00 7463
G 0.63 25.17 15.71 9874
F 0.34 11.11 3.73 18387
W 0.19 9.29 1.74 32914
O 0.52 9.36 4.85 11910
Unit 5.36 30.70 164.50 1151

Explanations: B – steam boiler and its auxiliaries, T – steam turbine and its auxiliaries,
G – generator and its auxiliaries, F – system of feed water pumps, W – system of cooling water
and service water pumps, O – others (including failures of electrical devices).

The results indicate that the evaporator is the most defective element in
technological track of 370 MW power unit. The mean time of the shutdown
duration is approx. 41 h, which – with the expected number of failures over
the year of 1.62 – gives the total duration of shutdowns of approx. 66 h. The
expected time between failures of 3820 h is, in this case, the shortest of all the
elements under investigation.
26 J. Buchta and A. Oziemski

Table 4. Forecasts for reliability indices of the most defective elements of the BB-1150 boiler.

Boiler element Expected Mean time of Total time of Mean time

failure rate a shutdown shutdowns between failures
[1/a] [h] [h/a] [h]
Economiser 0.40 37.76 14.97 15551
Evaporator 1.62 40.92 66.07 3819
Inner suspended 0.40 40.61 16.10 15619
pipes P1A
Convection su- 0.62 44.39 27.49 9959
perheater P1B
Outlet super- 0.26 51.32 13.56 23336
heater PIV
Reheater M1 0.35 39.75 14.05 17450
Reheater M2 0.18 46.20 8.22 34650
All superheaters 1.46 43.61 63.67 4224
(total)

Table 5. Reliability indices for dozen power units of Belchatow Power Station.

Failure Expected Mean time of Total time of Mean time

location failure rate a shutdown shutdowns between failures
[1/a] [h] [h/a] [h]
Unit 1 7.31 30.36 222.06 843
Unit 2 7.36 29.85 219.55 838
Unit 3 4.41 31.75 140.03 1398
Unit 4 5.26 30.49 160.30 1173
Unit 5 5.98 31.67 189.49 1030
Unit 6 5.50 29.72 163.40 1121
Unit 7 6.09 35.70 217.39 1012
Unit 8 5.86 32.06 187.84 1052
Unit 9 4.83 32.14 155.07 1278
Unit 10 4.97 24.18 120.06 1241
Unit 11 3.88 28.87 111.95 1590
Unit 12 4.41 34.16 150.73 1398

5 Conclusions
Power system safety is determined by its weakest links related to large power
units components. Optimal scheduling of the power unit maintenance, repairs
and modernization becomes impossible without knowledge of failure reasons and
Probabilistic methods in the reliability assessment. . . 27

Table 6. Reliability indices for BB-1150 steam boilers installed in Bełchatów Power Station.

Failure Expected Mean time of Total time of Mean time

location failure rate a shutdown shutdowns between failures
[1/a] [h] [h/a] [h]
Boiler 1 5.44 — 207.41 1134
Boiler 2 5.22 36.90 192.62 1181
Boiler 3 2.43 45.14 109.53 2541
Boiler 4 3.65 38.65 141.24 1687
Boiler 5 4.72 39.51 186.37 1307
Boiler 6 3.81 36.38 138.59 1618
Boiler 7 4.11 37.63 154.61 1500
Boiler 8 3.91 35.86 140.30 1576
Boiler 9 3.56 40.74 144.84 1734
Boiler 10 3.42 31.37 107.31 1802
Boiler 11 2.19 35.05 76.81 2814
Boiler 12 2.60 39.64 103.00 2373

failure rates as well as giving reasonable guarantees of continuity of power supply.

Reliability studies of power stations used to consider their weakest elements as an
integrated part of the unit, i.e. turbine and boiler together with their auxiliaries
and control systems. Consideration of the separate elements of the power unit
leads to complicated reliability models which are labour-consuming or hard to
analyze because of lack of information on failure rates of each element. Failures
of many auxiliary systems have no significant influence on power plant operation
but make it more difficult [2].
Reliability studies related to lignite fired power units rated at 370 MW and in-
stalled in domestic power system have been carried out systematically by authors
for a number of years. The results allow to analyze a changeability of reliability
indices in a long-term perspective and define reasons or effects of defective oper-
ation of generating devices. The knowledge of the current and forecast values of
failure rates can be useful to determine modernization schedules or to rationalize
repair and materials management. The failure rate is a reasonable measure for
a durability of generating devices and an indication for economical effectiveness
of repairs.
An acquisition of reliability indices proves that failure rate of the boiler makes
it the weakest link in power unit reliability as well boiler failure rate is crucial
for power plant availability. Therefore it is reasonable to reduce the amount of
shutdowns related with boiler failures by the way of focusing the maintenance
management on a lifetime improvement of the weakest links.
28 J. Buchta and A. Oziemski

Acknowledgements This work was financed by the Polish National Science

Centre under grant No. N513 346040.

Received in October 2011

References
[1] Allan R.N., Billinton R.: Power system reliability and its assessment. Power Engineering
J. July 1992, volume 6, issue 4, 191-196.
[2] Zadrzyński E.: Reliability. Informatics, control and management in power systems. PWN,
Warsaw 1979 (in Polish).
[3] Oziemski A., Sikora R.: Modeling of reliability of 370 MW units installed in BOT Bełchatów
Power Station. Wiadomości Elektrotechniczne 12(2006), 34–38 (in Polish).
[4] Oziemski A., Sikora R.: Modeling of reliability of 370 MW units installed in BOT Bełchatów
Power Station. In: Proc. XII Conference ZkwE’07 “Computer Applications in Electrical
Engineering”, Poznań, April 2007, 207–208 (in Polish).
[5] Buchta J., Oziemski A., Pawlik M.: Estimation of reliability measures of 370 MW lignite
fueled power units operating in Poland. In: Proc. 6th International Scientific and Technical
Conference “Efficiency and Power Quality of Electrical Supply of Industrial Enterprises”.
Mariupol (Ukraine), May 2008, 295–298.
[6] Buchta J., Oziemski A.: Reliability modeling of large coal-fired power units. In: Proc. 8th
International Conference “Control of Power Systems ’08”. Strbske Pleso (Slovakia), June
2008, CD proceeedings, 66.
[7] Oziemski A., Pawlik M.: Estimation of reliability measures for lignite fired 370 MW power
units. Kwartalnik Akademii Górniczo-Hutniczej w Krakowie, Górnictwo i Geoinżynieria
2(2009), 363–372 (in Polish).
[8] Barlow R.E., Proschan F.: Mathematical Theory of Reliability. Wiley, New York 1965.
[9] Lehmann E.L.: Testing Statistical Hypotheses. PWN, Warsaw 1968 (in Polish).
[10] Kececioglu D.: Reliability Engineering Handbook. DEStech Publ., 2002.
[11] Paska J., Wójcik P., Bargiel J., Goc W., Sowa P.: Data for electric power system calcula-
tions. In: Proc. 8th International Conference on Electrical Power Quality and Utilisation,
September 2005, 477–482.
[12] Lesiński S.: of investigation results by histogram method. Zeszyty Naukowe Politechniki
Łódzkiej ser. Elektryka 63(1978), 95–105 (in Polish).
[13] Pawlik M., Strzelczyk F.: Power Stations. WNT, Warsaw 2009 (in Polish).
Probabilistic methods in the reliability assessment. . . 29

Metody probabilistyczne w ocenie niezawodności bloków energetycznych

Streszczenie

W artykule dokonano oceny niezawodności pracy krajowych bloków 370 MW opalanych węglem
brunatnym, zainstalowanych w Elektrowni Bełchatów. W oparciu o koncepcję bloku reprezen-
tatywnego i metodę histogramu o założonej liczbie realizacji w klasach wyznaczono rozkłady
prawdopodobieństwa występowania czasów pracy i czasów awarii dla głównych urządzeń bloku
oraz ich najbardziej zawodnych elementów. Następnie na podstawie estymacji parametrów
uzyskanych w ten sposób rozkładów empirycznych określono dla tych elementów wartości oczeki-
wane ich podstawowych wskaźników niezawodnościowych: intensywności awarii, średniego czasu
wyłączenia, łącznego czasu wyłączeń w ciągu roku i czasu pracy bezawaryjnej.