Escolar Documentos
Profissional Documentos
Cultura Documentos
The reliability of mechanical systems and their parts decrease as they age
(Kumar, Jain, and Gandhi 2013), which increases catastrophic failures.
Hence, proper reliability estimation is required to plan an effective maintenance program of a repairable system or component in order to improve
operation time. Numerous researchers (Genschel and Meeker 2010; Kumar,
Jain, and Gandhi 2013; Olteanu and Freeman 2010; Zhang et al. 2013) used
the time-dependent parametric reliability model to analyze failure and repair
data of various nonrepairable or repairable systems in series, parallel, or combinations of series and parallel. The parametric reliability model assumes a
specic lifetime distribution; that is, exponential, Weibull, or lognormal,
where the operation time is considered as the only variable. But the reliability
of a system varies under different operating conditions (covariates), such as
temperature, humidity, dust, maintenance facilities, maintenance personal
skill, operator skill, etc. (Ghodrati 2005). The inuence of covariates (operating environments) may increase or decrease the reliability of a system. Poor
operating environments increase the failure rate as well as decrease reliability
and vice versa (Ghodrati 2005; Kumar and Klefsj
o 1994). Therefore, the
353
inuence of the external covariates needs to be considered along with the operation times (internal factors; Ahmad et al. 2011) for more reliable calculation
of system reliability.
Nonparametric reliability models have been developed based on the method proposed by Kaplan and
Meier (1958) and Nelson (1969) for consideration of
the inuencing covariates along with censored data
and the elimination of any specic distribution in
hazard analysis (Ghodrati 2005). The introduction of
a proportional hazard model (PHM) is a major contribution to the concept of nonparametric regression
methods. It was initially introduced in medical analysis by Cox (1972) and thereafter numerous researchers (Ahmad et al. 2011; Barabadi, Barabady, and
Markeset 2014; Ghasemi, Yacout, and Ouali 2007;
Ghodrati 2005; Ghodrati and Kumar 2005; Kumar
and Klefsj
o 1994; Kumar and Kelfsjo 1992; Lakshmi
and Sundari 2012; Li et al. 2007; Rigdon et al. 2013;
Tian and Liao 2011; Zhou, Chinnam, and Korostelev 2012) applied it in engineering for components
hazard analysis.
The literature survey reveals that there is a lack of
implementation of covariates-inuenced semiparametric reliability model in the failure analysis of a
coal-red boiler used in a thermal power plant.
Therefore, in this article, the probable covariates
that can inuence the failure rate are identied and
the application is demonstrated in the failure analysis. For this purpose, a Weibull proportional hazard
model (WPHM) was used.
[1]
.t; z1 / 0 .t/:exp.a:z1 /
D
D expfaz1 z2 g;
.t; z2 / 0 .t/:exp.a:z2 /
[2]
n
X
!) b1
ai zi
D u0 c b
1
[3]
iD1
1
MTBF0 D u0 :G 1 C
b
1
;
MTBFa D ua :G 1 C
b
[4]
[5]
time-dependent baseline characteristic life of a component, and G(x) is the gamma function.
According to Tutt et al. (2012), for the Weibull distribution of time between failure (TBF) data, the
expected number of failures at time t of a repairable system/component can be calculated as
Zt
N .t/ D
0
t b 1
t b
b:
dt D
:
u
u
[6]
[7]
[8]
The data used for the PHM of systems are the operation times and the values of covariates against each
operation time. The operation time may be uncensored
and/or censored. For a repairable system, the TBF may
be considered as uncensored data, whereas the operation time before overhauling may be considered as censored data. The parameters (a) are estimated by
maximizing the marginal, partial, or maximum likelihood functions (Kumar and Klefsj
o 1992). The partial
and marginal likelihood estimation gives the same
result for a. On the other hand, maximum likelihood
estimation of a is not effective for the model with a
large number of nuisance parameters. Therefore, the
marginal or partial likelihood estimation method is
widely used for estimation of a (Kumar and Klefsj
o
1992). Ties in the data set are handled by Breslow or
Efron approximations to the partial log-likelihood
(BIOST 515). If the value of a signicance test (p value)
is less than 0.05 (signicance level D 5 percent), then
the covariates have a signicant inuence on the failure
rate (Ahmad et al. 2011).
Identification of Covariates
The external covariates are identied from a brainstorming session between the expert operation and
Semiparametric Reliability Model
Case Study
The TBF (censored) and time before overhauling
(uncensored) data of the coal-red boiler tubes are collected from the maintenance logbook of a 250-MW
coal-red thermal power plant in the eastern region of
India. The tubes considered are FWTs, FSHTs, FRHTs,
PLSHTs, PSHTs, PRHTs, and economizer tubes
(ECOTs). First, the collected data were analyzed to
identify their independent and identical distribution
(i.i.d.) using trend tests and serial correlation tests
(Adhikary et al. 2012; Ghodrati 2005; Kumar and
Klefsj
o 1992). The tests reveal that all of the data are i.i.
d. For example, trend tests and serial correlation tests
of FWT are shown in the Figures 1 and 2, respectively.
Figure 1 shows that the data points cluster along a
straight line, representing no trend. Figure 2 shows that
the data points are scattered without any clear pattern, indicating no correlations. Hence, the data of FWT are i.i.d.
The Weibull distribution has been used to represent
equipment failure time because it is very exible and
the most commonly used model in failure data analysis
(Ghodrati and Kumar 2005). Kumar, Jain, and Gandhi
(2013) also supported that the Weibull distribution is
appropriate for increasingly degraded mechanical systems. Baseline parameters (u0 and b) of all of the boiler
tubes are estimated using the least-squares curve-tting
method (Adhikary et al. 2012; Kumar and Klefsj
o
1992). The estimated values are 9,160 hours and 2.04,
respectively (Table 1).
A brainstorming session was arranged with experienced operators, maintenance personnel, and managers
of the said power plant to assign values (qualitative) to
the considered covariates for each censored or uncensored operation time. Censored operation times are taken
as 0, whereas uncensored operation times are taken as 1.
The value of AP during an operation time is given as C1
when it is less than or equal to 10 percent; otherwise, 1.
1=2:04
1:478 AP 1:4
:
ua D 9160 exp
ST 0:612 OPSK
[10]
Significant covariates
FWT
FSHT
FRHT
PLSHT
PSHT
PRHT
ECOT
u0
ua
(MTBF)0
(MTBF)a
2.04
1.35
3.02
2.43
2.7
2.85
1.69
9,160
7,144
14,609
13,835
14,875
11,631
3,689
1,656
1,756
9,764
5,132
10,060
4,729
2,389
8,115
6,550
13,051
12,268
13,179
10,365
3,292
1,467
1,609
8,723
4,551
8,913
4,214
2,132
2:04 t 2:04 1
.t; z/ D
( 9160 9160
)
1:478 . 1/ 1:4
exp
:
. 1/ 0:612 . 1/
[11]
"
(
)# 1=2:04
1:478 . 1/ 1:4
ua D 9160 exp
. 1/ 0:612 . 1/
D 9160 0:180724 D 1655 .hr/:
[12]
In order to describe the proportionality of the hazard rates, h(t, z)high ash and h(t, z)low ash are the corresponding hazard rates assumed where the ash content
is greater than and less than the considerable limit
respectively. Then, putting the AP values of 1 and
C1 respectively in Eq. [11] and keeping the other
Semiparametric Reliability Model
N0(8760)
0.91
1.32
0.21
0.33
0.24
0.45
4.31
Na(8760)
29.95
8.76
0.72
3.67
0.69
5.79
8.99
hazard rates.
Table 1 shows that out of the ve external covariates, AP and ST are the most signicant covariates
because most of the subsystems are inuenced by
them. By considering the inuence of the external
covariates, the characteristic life (ua) and corresponding (MTBF)a of each tube are extensively reduced
and the corresponding probability of failure is
greater than the expected failures under parametric
reliability model.
CONCLUSIONS
In the present research work, ve different operating
environments (covariates) that inuence reliability are
considered for coal-red boiler tubes. The WPHM, a
semiparametric reliability model, is applied to estimate
their degree of inuence. The goodness-of-t test shows
that the PHM assumption of the failure data is correct.
The case study reveals that most of the tubes are signicantly affected by AP and ST compared to the other
three covariates. Hence, the inuence of these two covariates is more signicant for consistent reliability calculation. The characteristic life (ua) and the MTBF are
drastically reduced when considering the inuence of
the operating environments. Preventive maintenance
planning along with corresponding spare parts provisioning based on these reduced characteristic life (ua)
and MTBF will be effective. This study will help the
industry to plan preventive maintenance planning and
procurement of spare parts in a cost-effective manner.
358
Souren Mitra is a professor and Head of the Production Engineering Department at Jadavpur University,
Kolkata, India. His research areas include reliability
and maintenance engineering, nontraditional machining process, nanomachining, and advanced
manufacturing systems. He has published more than
50 research papers in international and national journals and conferences. Several Ph.D. and master of engineering theses have been completed under his
guidance. He is a recipient of the Research Award of
the University Grand Commission, New Delhi, and
has completed three sponsored research projects as the
principal investigator. Dr. Mitra is a member of the
Institute of Engineers (India). He has attended several
international conferences abroad.
REFERENCES
Adhikary, D. D., G. K. Bose, D. Bose, and S. Mitra. 2014. Multi criteria
FMECA for coal-fired thermal power plants using COPRAS-G. International Journal of Quality & Reliability Management 31(5): 601
14. http://dx.doi.org/10.1108/IJQRM-04-2013-0068
Adhikary, D. D., G. K. Bose, S. Chattopadhyay, D. Bose, and S. Mitra.
2012. RAM investigation of coal-fired thermal power plants: a case
study. International Journal of Industrial Engineering Computations
3: 42334. http://dx.doi.org/10.5267/j.ijiec.2011.12.003
Ahmad, R., S. Kamaruddin, I. Azid, and I. Almanar. 2011. Maintenance
management decision model for preventive maintenance strategy
on production equipment. Journal of Industrial Engineering Internationa 7(13): 2234.
Barabadi, A., J. Barabady, and T. Markeset. 2014. Application of reliability models with covariates in spare part prediction and optimizationa case study. Reliability Engineering and System Safety
123: 17. http://dx.doi.org/10.1016/j.ress.2013.09.012
BIOST 515, (2004). Staff Lecture 17, March 4. Cox proportional hazards
models. University of Washington, Seattle. http://courses.washington.edu/b515/l17.pdf.
Cox, D. R. 1972. Regression models and life-tables. Journal of the Royal
Statistical Society B34: 187220.
Ebeling, C. E. 2008. An introduction to reliability and maintainability
engineering. 10th ed. New Delhi, India: Tata McGraw-Hill.
Genschel, U., and W. Q. Meeker. 2010. A comparison of maximum likelihood and median-rank regression for Weibull estimation. Quality
Engineering
22:
23655.
http://dx.doi.org/10.1080/
08982112.2010.503447
Ghasemi, A., S. Yacout, and M. S. Ouali. 2007. Optimal condition based
maintenance with imperfect information and the proportional hazards model. International Journal of Production Research 45(4):
9891012. http://dx.doi.org/10.1080/00207540600596882
359
APPENDIX
TABLE A1 Operation Times of FWT and Codified Values of Influencing Covariates
TBF
Censored/uncensored
10,580
7,640
8,600
5,900
10,500
12,500
2,744
11,700
12,400
8,960
4,793
12,400
2,970
3,200
9,200
4,500
10,918
5,597
5,045
6,500
2,300
17,000
1
0
1
1
0
1
1
0
1
0
1
0
1
1
1
0
1
0
1
1
0
0
AP
ST
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
pH
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
OPSK
MPSK
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
1
TABLE A2 Regression Coefficients (a) of FWT (Parameters/Coefficients Are Estimated by Maximizing the Partial Likelihood Function)a
Coefficients (a)
of the covariates
Estimated
value
AP
ST
OPSK
1.478
1.400
0.612
Standard
error
Walds
chi-square
Pr > chi2
Hazard
ratio
Hazard ratio
lower bound (95%)
Hazard ratio
upper bound (95%)
0.597
0.526
0.435
6.135
7.097
1.980
0.013
0.008
0.159
0.228
0.247
0.542
0.071
0.088
0.231
0.735
0.691
1.272
Significance level: 5%. Ties handling: Breslows method. Stop conditions: iterations: 100; convergence: 0.0001. Model selection: backward.
360