Content uploaded by Sanjay Chaturvedi
Author content
All content in this area was uploaded by Sanjay Chaturvedi on Oct 20, 2015
Content may be subject to copyright.
International Journal of Quality & Reliability Management
Prioritization of maintenance tasks on industrial equipment for reliability: A fuzzy
approach
Edwin Vijay Kumar S.K. Chaturvedi
Article information:
To cite this document:
Edwin Vijay Kumar S.K. Chaturvedi, (2011),"Prioritization of maintenance tasks on industrial equipment for
reliability", International Journal of Quality & Reliability Management, Vol. 28 Iss 1 pp. 109 - 126
Permanent link to this document:
http://dx.doi.org/10.1108/02656711111097571
Downloaded on: 19 October 2015, At: 23:24 (PT)
References: this document contains references to 21 other documents.
To copy this document: permissions@emeraldinsight.com
The fulltext of this document has been downloaded 1095 times since 2011*
Users who downloaded this article also downloaded:
Amik Garg, S.G. Deshmukh, (2006),"Maintenance management: literature review and
directions", Journal of Quality in Maintenance Engineering, Vol. 12 Iss 3 pp. 205-238 http://
dx.doi.org/10.1108/13552510610685075
A. Crespo Márquez, P. Moreu de León, J.F. Gómez Fernández, C. Parra Márquez, M. López
Campos, (2009),"The maintenance management framework: A practical view to maintenance
management", Journal of Quality in Maintenance Engineering, Vol. 15 Iss 2 pp. 167-178 http://
dx.doi.org/10.1108/13552510910961110
Sephooko I. Motelle, Nicholas Biekpe, (2014),"Financial intermediation spread and stability of the banking
system in the Southern Africa Customs Union", Managerial Finance, Vol. 40 Iss 3 pp. 276-299 http://
dx.doi.org/10.1108/MF-06-2013-0147
Access to this document was granted through an Emerald subscription provided by emerald-srm:395687 []
For Authors
If you would like to write for this, or any other Emerald publication, then please use our Emerald for
Authors service information about how to choose which publication to write for and submission guidelines
are available for all. Please visit www.emeraldinsight.com/authors for more information.
About Emerald www.emeraldinsight.com
Emerald is a global publisher linking research and practice to the benefit of society. The company
manages a portfolio of more than 290 journals and over 2,350 books and book series volumes, as well as
providing an extensive range of online products and additional customer resources and services.
Emerald is both COUNTER 4 and TRANSFER compliant. The organization is a partner of the Committee
on Publication Ethics (COPE) and also works with Portico and the LOCKSS initiative for digital archive
preservation.
*Related content and download information correct at time of download.
Downloaded by Indian Institute of Technology Kharagpur At 23:24 19 October 2015 (PT)
Prioritization of maintenance
tasks on industrial equipment
for reliability
A fuzzy approach
Edwin Vijay Kumar and S.K. Chaturvedi
Reliability Engineering Centre, Indian Institute of Technology, Kharagpur, India
Abstract
Purpose – This paper aims to prioritize preventive maintenance actions on process equipment by
evaluating the risk associated with failure modes using predictive maintenance data instead of
maintenance history alone.
Design/methodology/approach – In process plants, maintenance task identification is based on
the failure mode and effect analysis (FMEA). To eliminate or mitigate risk caused by failure modes,
maintenance tasks need to be prioritized. Risk priority number (RPN) can be used to rank the risk. RPN
is estimated invariably using maintenance history. However, maintenance history has deficiencies,
like limited data, inconsistency etc. To overcome these deficiencies, the proposed approach uses the
predictive maintenance data clubbed with expert domain knowledge. Unlike the traditional single step
approach, RPN is estimated in two steps, i.e. Step 1 estimates the “Possibility of failure mode
detection” and Step 2 estimates RPN using output of step 1. Fuzzy sets and approximate reasoning are
used to handle the uncertainty/imprecision in data and subjectivity/vagueness of expert domain
knowledge. Fuzzy inference system is developed using MATLABw6.5.
Findings – The proposed approach is applied to a large gearbox in an integrated steel plant. The
gearbox is covered under a predictive maintenance program. RPN for each of the failure modes is
estimated with the proposed approach and compared with the maintenance task schedule. The
illustrative case study results show that the proposed approach helps in detection of failure modes
more scientifically and prevents “Over maintenance” to ensure reliability.
Originality/value – This approach gives an opportunity to integrate the predictive maintenance
data and subjective/qualitative domain expertise to evaluate the possibility of failure mode detection
(POD) quantitatively, which is otherwise purely estimated using subjective judgments. The approach
is generic and can be applied to a variety of process equipment to ensure reliability through prioritized
maintenance scheduling.
Keywords Fuzzy logic, Maintenance reliability, Risk analysis
Paper type Research paper
1. Introduction
In the recent past, reliability centered maintenance (RCM) strategy has gained more
adaptability in process industry due to its strength in deciding maintenance
requirements using a structured logical approach to ensure high levels of reliability at
optimum maintenance expenditure (Nowlan et al., 1978; Moubray, 1997). With
reference to industries such as steel, power and oil, it is observed that 3 per cent-5 per
cent of the turnover is spent on maintenance. Pintelon et al. (1999) addressed the
problem of designing an optimum maintenance concept for process equipment (paint
spraying robot) using the concept of RCM. The process plant equipment needs more
specialized approach in deciding maintenance requirements, since operating context
The current issue and full text archive of this journal is available at
www.emeraldinsight.com/0265-671X.htm
Prioritization of
maintenance
tasks
109
Received January 2008
Revised June 2008
Accepted October 2008
International Journal of Quality &
Reliability Management
Vol. 28 No. 1, 2011
pp. 109-126
qEmerald Group Publishing Limited
0265-671X
DOI 10.1108/02656711111097571
Downloaded by Indian Institute of Technology Kharagpur At 23:24 19 October 2015 (PT)
plays a major role in the degradation process. Operating context is a set of operating
conditions in which equipment operates. Deshpande
´and Modak (2002) have applied
RCM on steel rolling mill to develop a maintenance strategy by addressing equipment
specific needs, but the maintenance tasks are not prioritized. RCM frame work does not
offer the approach to do the same. Failure mode effect analysis (FMEA) can be utilized
to identify and prioritize the maintenance tasks based on failure modes using either a
Criticality Index or Risk Priority Number (RPN).
Sharma et al.(2006) proposed an approach using fuzzy inference system (FIS) with
FMEA to compute RPN with maintenance history. In their approach, the parameters to
compute RPN namely, frequency of occurrence, severity of risk and non detectability of
failure mode, are considered as fuzzy sets using appropriate membership functions.
The membership functions for the parameters are deduced from the maintenance
records integrated with maintenance expert’s knowledge. The rule base is generated to
apply approximate reasoning on input variables at various levels. The fuzzy
conclusion is aggregated and defuzzified to get a crisp value for RPN.
The well-known RPN expression is given in (1):
RPN ¼Of£S£Dp;ð1Þ
where O
f
¼frequency of failure mode occurrence, S¼severity of failure effect and
Dp ¼probability of non detection of failure mode. Hence, RPN is more suitable index
for prioritization of maintenance tasks in process plants.
While computing RPN (Sharma et al.(2006)), the non detectability of the failure mode
occurrence is represented by fuzzy sets using the domain knowledge and classified in
terms of linguistic variables, such as low, remote, highly possible, etc. Score is awarded
to the classification based on the subjective judgment by the domain experts or limited
maintenance data. The approach discussed so far provides, a solution in the absence of
adequate maintenance information. But, when the equipment is monitored under
predictive monitoring (PDM), useful information is generated in the form of
quantitative level of condition indicating parameters. Moreover, other associated PDM
information i.e. operating context and frequency of monitoring could also be used to
estimate possibility of failure mode detection (POD) along with condition monitoring
data. Place et al. (2000) had used the predictive monitoring data of a helicopter gearbox
to model degradation process with Bayesian approach and used the same to find out
the dependency of the POD to detect functional failure. The model had estimated POD
probabilistically and brought out the importance of probability of failure detection as
expressed in (2) before the potential failure matures as a functional failure:
pðFFÞ¼pðFM Þ:ð12pðPODÞÞ ð2Þ
where p(FF) ¼probability of functional failure, p(FM) ¼probability of occurrence of
failure mode and p(POD) ¼probability of detection of failure mode.
To integrate the maintenance expert’s domain knowledge with PDM data, fuzzy
approach is more suitable for handling both quantitative and qualitative data.
Therefore, the present paper extends the above two approaches in two ways, i.e.:
(1) proposes to use PDM data in place of maintenance history; and
(2) estimating POD with quantitative data from PDM instead of purely relying on
expert’s opinion.
IJQRM
28,1
110
Downloaded by Indian Institute of Technology Kharagpur At 23:24 19 October 2015 (PT)
The PDM data are collected and recorded quantitatively using different gadgets, like
vibration recorders, thermal imagers, flow indicators etc. The proposed approach is
generic and computes RPN two steps using fuzzy inference system (FIS). The approach
is demonstrated with application to a large gearbox of an integrated steel plant and
results are presented.
The paper is organized in five sections. Section 1.1 deals with necessity of RCM and
FMEA-RPN approach in process plants. Section 2.0 outlines the fuzzy inference system
to use with FMEA. Section 3.0 details the proposed approach with the use of predictive
maintenance data to estimate possibility of failure mode detection and RPN. Section 4.0
validates the approach with practical case study. Section 5.0 is summary and conclusions.
1.1 Failure mode effect analysis – risk assessment – priority of maintenance task
RCM primarily addresses the maintenance requirements of the equipment with an
emphasis on functional failures and failure modes along with operating context. While
enlisting the functional failures, FMEA is carried out thoroughly on the system and
sub-systems to ensure the identification of all possible functional failures with
associated failure modes.
FMEA is structured inductive approach (Misra, 1992), wherein a system is sub
divided into its number of constituent parts based on their functional complexity and
sub-functional relationships as shown in Figure 1. Functional sub-functional division
mainly depends on the processes and motive of risk/reliability assessment.
In process industries, sub-functional break up is done in such a way that the
resultant failure cause can be eliminated or mitigated through a manageable
maintenance activity on a specific component. For complex equipment, there may be
numerous preventive tasks, which need to be carefully prioritized in order to ensure
both process reliability and equipment reliability. To plan a preventive maintenance
task, the detection of failure mode becomes vital. Otherwise entire preventive planning
would go in waste.
Figure 1.
Function and
sub-functions bottom up
relation
Prioritization of
maintenance
tasks
111
Downloaded by Indian Institute of Technology Kharagpur At 23:24 19 October 2015 (PT)
2. Process plant equipment – failure mode detection
Process equipment can be classified into three categories, critical, semi-critical and
non-critical based on process criticality, operating context and complexity of the
equipment. The equipment degradation is estimated using condition indicators (or
variables) like, vibration, temperature etc. The amplitude levels of the condition
indicators reflect the level of degradation, i.e. higher the level, higher would be the
degradation but easier to detect abnormality.
The operating context is another important factor, which influences degradation.
Operating context is classified into three operating regimes (Edwin and Chaturvedi,
2006), namely, normal, marginal and hostile operating contexts as reproduced in
Table I. Under the influence of operating context, it is practically found that condition
indicator levels are modulated, thereby true level of degradation can not be inferred by
the mere trending of condition indicator’s level. Hence, operating context needs to be
considered while using condition indicator level to infer state of equipment. Invariably,
the maintenance domain expert does not have a precise knowledge about the
correlation of the operating context and level of degradation for a given level of
condition indicator on the specific equipment (or group of equipments).
2.1 Potential – functional failure and probability of failure mode detection
The frequency of monitoring is decided based on the potential-functional failure
interval (PF Interval). Other factors like, operating context, method of monitoring, rate
of degradation and acceptability criteria etc., are also considered while fixing the
monitoring frequency (monitoring schedule). Equipment operating under hostile
operating context needs frequent monitoring than the equipment working under
normal operating context. The identification of precise frequency of monitoring is
somewhat difficult, as the P-F interval itself cannot be decided with equipment
specifications. The uncertainty between initiation of degradation and reaching to
unacceptable levels is shown in Figure 2.
As shown in Figure 2, the performance decreases over a period of time and drops
down to an unacceptable level (potential failure) and leads to a functional failure at a
later time. The degradation of a component/system may vary widely due to
abovementioned factors. Owing to this variation and associated randomness, the
potential and functional failure times are also not precise. Hence, the frequency of
monitoring decided on the basis of PF interval (as given by (3)) is also a variable and
some applied researchers say PF interval itself is elusive (Murray, 2007):
PF Interval ¼tf2tpð3Þ
Parameter Normal Marginal Hostile
Relative humidity% ,80.0 80.0-90.0 .90.0
Dust Nil Traces High
Ambient temp (8C) ,40.0 40.0-50.0 .50.0
Duty cycle Constant Varying Shock
Vibration/shock from surrounding Nil Mild Severe
Installation Indoor Semi-outdoor Outdoor
Accessibility for maintenance Good Restricted Nil
Table I.
Classification of
operating context in
process industries
IJQRM
28,1
112
Downloaded by Indian Institute of Technology Kharagpur At 23:24 19 October 2015 (PT)
where t
f
and t
p
are time for functional failure and potential failures. Since this interval
is uncertain and degradation is a random event, predicting an impending failure is
highly probabilistic (Stewart and Melchers, 1997). Therefore, the possibility of
detecting a failure in advance is not a crisp event and fuzziness is associated with it.
This uncertainty can be better handled with fuzzy logic to arrive at estimating
appropriate possibility level instead of probability due to scarcity of required inputs.
2.2 Possibility of failure mode detection
Estimating the failure probability distribution with limited/un-organized maintenance
data is cumbersome and not rewarding. Possibility distribution of fuzzy sets (Zadeh,
1975) and transformation of probability distributions to possibility distributions
(Dubois and Prade, 1986) and vice-versa offers a great flexibility and simplicity to
compute possibility of events. The strength of the fuzzy logic in the field of reliability
engineering is well documented by Bowles and Pelaez (1995).
The efficiency of PDM program can be quantified by the ratio of failure modes
detected in time to the total number of failure modes occurred in a specified interval of
time on specified equipment. This approach proposes to estimate the “Possibility of
failure mode detection” (POD) using three fuzzy variables in FIS with a suitable
membership functions for each variable, i.e.:
(1) level of condition indicator;
(2) frequency of monitoring; and
(3) operating context.
These variables can be expressed as fuzzy sets with a linguistic description as given in
Table II. For instance, treating “condition indicator level” as fuzzy variable, it can be
expressed the membership function shown in Figure 3.
Fuzzy inference system can be divided into four building-blocks, such as:
(1) fuzzifier;
(2) rule base;
(3) inference engine; and
(4) defuzzifier.
Figure 2.
Uncertainty in PF interval
Prioritization of
maintenance
tasks
113
Downloaded by Indian Institute of Technology Kharagpur At 23:24 19 October 2015 (PT)
A schematic representation of fuzzy inference system used in this approach is shown in
Figure 4 with a brief description as follows:
Input variable is fuzzified using the fuzzy linguistic variable, like low, medium and
high etc, with an appropriate membership functions. The selection of membership
function is subjective and depends on the data type and its variability in the domain.
Expert rule base is, a set of rules mapping the inputs to output under various input
levels. There can be many rules, which are applicable simultaneously on the input
variables and leading to many similar or dissimilar output situations. Aggregating
these outputs can be done using a method popularly known as centroid method of
aggregation or Mamdani rule, which deals with Max-Min approach (Ross, 1997). With
the reliability perspective, when the POD is high, the reliability can be kept at higher
levels as the required maintenance action to contain degradation level can be planned
in advance, before an impending failure matures as a failure. With the risk perspective,
risk can be minimized as mitigation can effectively be planned (Jardine, 2002). A
typical POD values computed using above mentioned approach is presented in
Table III. More details are presented in the case study.
3. Predictive maintenance data and fuzzy estimation of RPN: approach
The estimation of RPN in two stages as proposed in this paper heavily relies on the
predictive maintenance data. The predictive maintenance data can be generated during
the periodical monitoring of the industrial equipment with the help of online and off-line
measurements, and data recording systems. Online monitoring systems are permanently
installed measuring, transmitting and recording devices, such as vibration, temperature,
Figure 4.
Elements of fuzzy
inference system
Figure 3.
Fuzzy sets for condition
indicator level
(normalized)
Fuzzy variable Fuzzy levels Membership function
Condition indicator level High, medium, low Triangular
Frequency of monitoring Very frequent, frequent, rare Triangular
Operating context Hostile, marginal, normal Trapezoidal
Table II.
Fuzzy set details for
estimating POD with
PDM data
IJQRM
28,1
114
Downloaded by Indian Institute of Technology Kharagpur At 23:24 19 October 2015 (PT)
flow, pressure monitors etc, and off-line instruments are like handheld vibration data
collectors, thermal imagers, ultrasonic leak/crack detectors etc.
The RPN as discussed in the Section 1 and computed using (1), the estimation of the
RPN requires, three quantities, i.e. Frequency of Occurrence of failure mode (O
f
),
Severity of risk (S) and probability of non detection of failure mode (D
p
). Since the
present approach uses fuzzy estimation, the probability of non detection of failure
mode is replaced with possibility of failure mode detection (POD) represented by fuzzy
sets. Other two variables are also represented by fuzzy sets. In the first step of the
approach, the possibility of failure mode detection can be estimated using the fuzzy
input variables and FIS-I and the output of the FIS-I can be used as input to FIS-II
along with other two inputs to estimate RPN. Table IV provides fuzzy sets and their
corresponding FIS input – output relationships.
The frequency of occurrence of the failure mode (O
f
)is found from the number of
withdrawals of equipment from the service in a planned way to eliminate the specified
failure mode. This is possible by trending the condition indicator level over a period of
time. As the trend level reaches the unacceptable level, the equipment is withdrawn
from the service to replace/repair the component, so that the failure mode is eliminated.
Condition indicator level Frequency of monitoring Operating context
Possibility of detection
(POD)
0.089 (Low) 0.0299 (Rare) 0.526 (Marginal) 0.314 (Much less possible)
0.445 (Marginal) –do– –do– 0.533 (Less possible)
–do– –do– 0.893 (Hostile) 0.097 (Not possible)
–do– 0.859 (Very frequent) –do– 0.682 (Possible)
0.792 (High) –do – 0.201(Normal) 0.865 (Highly possible)
Table III.
Variation of possibility of
detection
Input variable Data required
Fuzzy set/
Membership
function FIS
Output fuzzy set/
Membership
function
Step 1
Condition indicator
level
Condition indicator value Triangular FIS
–I
Possibility of
detection
Triangular five
levels
Frequency of
monitoring
Schedule of monitoring Triangular FIS
–I
Operating context Operating conditions Trapezoidal FIS
–I
Step 2
Frequency of
occurrence of failure
mode
Withdrawal of equipment for
maintenance on a specified failure
mode
Triangular FIS
–II
RPN
Severity of risk Expected damage in the absence of
preventive maintenance action
Triangular FIS
–II
Triangular
Possibility of
detection
FIS – I output Triangular FIS
–II
Five levels
Table IV.
Details of fuzzy sets for
estimation of RPN
Prioritization of
maintenance
tasks
115
Downloaded by Indian Institute of Technology Kharagpur At 23:24 19 October 2015 (PT)
The best example is that, when the equipment bearing temperature rises to an
unacceptable level, the most probable failure mode would be the bearing failure
(seizure or break). To avert this failure mode, the bearing would be thoroughly checked
and replaced (if necessary), otherwise necessary maintenance actions like adjustment
of bearing clearance, alignment etc will be done.
The severity of the risk can be estimated from the equipment configuration and
operational criticality. The failure mode can be traced back to the component level, and
the consequence due to damaged component can be used to estimate the severity in the
linguistic form, like very low, moderate and high.
Case study
To demonstrate the usefulness of the above approach, predictive maintenance data on
the large gearbox, which drives the steel rolling mill stand, is used. The gearbox
specifications are as given in Table V.
The FEMA along with condition indicators is as shown in the Appendix (Table AI).
Five failure modes are identified with the gearbox, namely, abnormal sound (F1), abnormal
vibration (F2), bearing temperature (F3), bearing vibration (F4) and shaft shear (F5).
Step 1
The relationship between failure modes and PDM data is shown in the Table VI to
estimate possibility of failure mode detection (POD) of the gearbox. The condition
indicator level is normalized and fuzzified to express the transition in linguistic
variables (refer to Figure 3). Other two variables, frequency of monitoring and
operating context are also fuzzified.
FIS – I named as “FIS – possibility” is used. The simple logic with expert opinion is
applied to form rules to develop FIS-I with the above fuzzy variables at different levels.
A total of 27 rules are formed with three input fuzzy variables, each with three levels.
Using these three input variables to FIS, the output variable “Possibility of detection”
is estimated in five levels, namely, “not possible”(NP), “very less possible”(VLP), “less
possible”(LP), “possible”(P) and “highly possible”(HP). The outcome in five levels is
Failure mode Indicator level Frequency of monitoring Operating context
F1 (Abnormal sound) Body vibrations Very frequent Hostile
F2 (Abnormal vibration) Oil flow Frequent Hostile
Oil temperature Very frequent Hostile
F3 (Bearing temperature) Bearing temperature Very frequent Hostile
F4 (Bearing vibration) Shock pulse rate Frequent Hostile
F5 (Shaft shear) Nil Rare Hostile
Table VI.
Relationship of failure
mode – predictive
monitoring
Input speed: 440-880 RPM
Gearbox type: Bevel gear
Normal output speed: 368.35
RPM Normal output torque: 24.36 kN-M
Gear ratio: 1.5
Gear teeth: 22/33 Weight: 2,805 kgs
Oil pressure: Min. 1.2
Bar, Max. 1.8 Bar Operating temperature: 38.0 deg C to 53.0 deg C
Table V.
Gearbox specifications
IJQRM
28,1
116
Downloaded by Indian Institute of Technology Kharagpur At 23:24 19 October 2015 (PT)
useful to grade possibility with a smooth transition from one level to another. The
triangular membership functions for possibility of detection are as shown in Figure 5.
A snapshot of rule base and the aggregated output are shown in Figure 6. The
output derived using the max-min method is a fuzzy set and needs to be defuzzified to
obtain a crisp value for further use. For example, if the indicator level is “low” and
frequency of monitoring is “rare” and operating context is “marginal” the rule base
concludes the possibility of detection is “very less possible”. This rule base is generic
and not equipment/plant specific and customization of the rule base to a specific
application is simple.
Table VII shows the variation of possibility of detection of failure mode with the
variation in indicator level, frequency of monitoring and operating context.
Quantitative fuzzy output is shown in parenthesis for various combinations of input
variables. The “surface viewer” facility available in Matlabwcan also be utilized to
have a pictorial view of variations in the output for a combination of input variables in
the defined range. Once the possibility of detection is estimated with dynamic data, the
same is used to compute RPN using FIS-II in a second step as discussed in the
following section.
Figure 6.
Firing of rules on inputs –
aggregation of output
(POD)
Figure 5.
Fuzzy sets for possibility
of failure mode detection
Prioritization of
maintenance
tasks
117
Downloaded by Indian Institute of Technology Kharagpur At 23:24 19 October 2015 (PT)
Condition
indicator level POD
Failure mode Indicator units, allowable level Initial Final
Indicator
normalized
value Operating context
Frequency of
monitoring Initial Final
F1: Abnormal sound Body vibrations (mm/sec), 11.2 3.5 8.9 0.312 0.794 Hostile VF 0.886 0.889
F2: Abnormal vibration Oil flow (lpm), 55.0 40.0 40.0 0.727 0.727 Hostile F 0.734 0734
Oil temperature (8C), 62.0 53.0 57.0 0.854 0.919 Hostile VF 0.889 0.892
F3: Bearing temperature Bearing temperature (8C), 75.0 67.0 73.0 0.89 0.97 Hostile VF 0.891 0.893
F4: Bearing vibration Shock pulse rate (no.), 60 34 49 0.56 0.81 Hostile F 0.723 0.733
F5: Shaft shear Nil – – 0.05 0.05 Hostile R 0.535 0.535
Table VII.
Results of
“FIS-possibility of
detection” on gearbox
IJQRM
28,1
118
Downloaded by Indian Institute of Technology Kharagpur At 23:24 19 October 2015 (PT)
Step 2
The FIS-II named as “FIS-RPN” is used to estimate the RPN for the five failure modes
of the gearbox. A rule base is formed consisting of 125 rules with various possible
combinations of input variables given in Table VIII.
For instance, the computed RPN (output) both in linguistic form “High” is also
indicated with a crisp value as 0.645, for Of¼0:825, POD ¼0:645 and S ¼0:717. A part
of the developed expert rule base aggregated with inference engine is shown in Figure 7.
The fuzzy sets of input, frequency of occurrence (O
f
), and output, RPN, are shown in
Figure 8. Table IX presents the computed results where the crisp value of RPN is
indicated with a numerical value and rank is shown in parenthesis. From the results
shown in Table VII and Table IX, it can be seen that the PDM data is useful to compute
POD and the same is used to estimate RPN for the gearbox. The RPN clearly indicates
the changed priority of the maintenance tasks on the gearbox.
The approach is compared with the method proposed by Sharma et al.(2006), which
estimates the probability of non detection of failure mode from maintenance data in
Failure Mode Frequency of occurrence Severity Possibility of detection (linguistic)
F1 (Abnormal sound) High Very high Highly possible
F2 (Abnormal vibration) Very high Marginal Highly possible
Very high Marginal Possible
F3 (Bearing temperature) High Marginal Possible
F4 (Bearing vibration) Rare High Possible
F5 (Shaft shear) Very rare Very High Not possible
Table VIII.
Failure modes and fuzzy
sets of input variable of
FIS – II
Figure 7.
Firing of rubes – output
aggregation with
MATLAB
Prioritization of
maintenance
tasks
119
Downloaded by Indian Institute of Technology Kharagpur At 23:24 19 October 2015 (PT)
linguistic class of five fuzzy sets. A remarkable deviation is found when the POD is
computed with PDM data as shown in the Appendix (Table AII).
From the Table IX, it can be further observed that order of priority of the predictive
maintenance task changes after considering the predictive monitoring data. In normal
practice, as the indicator level approaches an alarm level, maintenance engineer
considers, that particular failure mode is more important than others. But, due to
influence of the operating context, actual levels may not be reflected for other
parameters, and it may also lead to failure due to a hidden failure mode.
For example, failure mode, Shaft shear (F5) priority is changed from order 5 to order
2 after considering the POD. In reality, it is very practical to check the healthiness of
the shaft for the equipment running under hostile operating conditions with fluctuating
loads. Otherwise failure of the shaft would be hidden and catastrophic (Flutter, 1995).
Possibility of detection
(POD) RPN
Failure mode
Frequency of
occurrence (O
f
) Severity (S) Linguistic Evaluated Initial Final
F1: Abnormal sound High Very high High possible 0.889 0.623 0.723
(1) (1)
F2: Abnormal vibration Very high Marginal High possible 0.734 0.474 (0.474)
(2)
Possible 0.892 0.474 0.474
(2) (3)
F3: Bearing temperature High Marginal Possible 0.893 0.403 0.403
(3) (4)
F4: Bearing vibration Rare High Possible 0.733 0.345 (0.345)
(4) (5)
F5: Shaft shear Very rare Very high Not possible 0.535 0.35 0.595
(5) (2)
Table IX.
Results of FIS-RPN on
gearbox failure modes
Figure 8.
Membership functions of
frequency of occurrence
and RPN
IJQRM
28,1
120
Downloaded by Indian Institute of Technology Kharagpur At 23:24 19 October 2015 (PT)
Therefore, as documented by Jardine et al. (2006), there is a tremendous need to
integrate the PDM data into prognostic process in an industry by developing suitable
models, frameworks which use soft computing tools and domain knowledge and the
same is). An attempt has been made in this direction by the authors. In a nutshell,
advantages of the proposed approach can be summarized as:
.Helps in quick review of ranking of numerous maintenance tasks.
.Removes ambiguity in maintenance planning.
.A tool to find out hidden failure causes with POD.
.Helps in eliminating “Over-maintenance”.
.Approach is component-failure mode specific.
.Highly flexible in dealing with uncertain data, expert opinions.
.Easy to customize.
5. Summary and conclusions
Maintenance needs of the process plants equipments can be better addressed with
RCM Approach. Failure mode and effect analysis, helps in identifying all possible
failure cause with a specific reference to the component of systems and sub-systems.
Predictive monitoring is not only a better maintenance strategy, but also helps in
achieving higher maintenance reliability.
In this paper, the approach is presented for computing RPN with integration of
RCM, FMEA, PDM and FIS to draw a better maintenance plan on industrial
equipment, which helps in mapping field data and maintenance engineer’s expertise.
Through the Aggregation of various variables like operating conditions, severity of
risk etc. (which are fuzzy in nature), have been taken care by framing an expert rule
base thereby reducing the dependency on equipment history. This approach presented
in this paper is generic in nature and can be used on any process equipment to guide
maintenance engineer to reach the objective with an engineering reasoning.
Further, this approach also offers a quick tool to access maintenance requirements
periodically for large number of equipments under various operating context and
predictive strategies. Fine tuning of the system is possible and full pledged fuzzy
inference system on a hardware platform can be developed for a real-time application.
The system is put in a pilot study in an integrated steel plant and observed
encouraging results.
References
Bowles, J.B. and Pelaez, C.E. (1995), “Application of fuzzy logic to reliability engineering”, IEEE,
Vol. 83 No. 3, pp. 435-49.
Deshpande
´, V.S. and Modak, J.P. (2002), “Maintenance strategy for tilting table of rolling mill
based on reliability considerations”, Reliability Engineering and System Safety, Vol. 80
No. 1, pp. 1-18.
Dubois, D. and Prade, H. (1986), “A set-theoretic view on belief functions: logical operations and
approximations by fuzzy sets”, International Journal of General Systems, Vol. 12,
pp. 193-226.
Edwin, V.K. and Chaturvedi, S.K. (2006), “Application of predictive maintenance techniques for
failure rate modelling of electric motors in process industry”, Proceedings of INCRESE06,
International Conference on Reliability and Safety Engineering 2006.
Prioritization of
maintenance
tasks
121
Downloaded by Indian Institute of Technology Kharagpur At 23:24 19 October 2015 (PT)
Flutter, D.N. (1995), “Vibration monitoring of industrial gearboxes using time domain averaging:
gearbox noise, vibration and diagnostics”, I Mech E.
Jardine, A.K.S. (2002), “Optimizing condition based maintenance decision”, IEEE Proceedings
Annual Reliability and Maintainability Symposium.
Jardine, A.K.S., Lin, D. and Banjevic, D. (2006), “A review on machinery diagnostics and
prognostics implementing condition-based maintenance”, Mechanical Systems and Signal
Processing, Vol. 20 No. 7, pp. 1483-510.
Misra, K.B. (1992), Reliability Analysis and Prediction: A Methodology Oriented Treatment,
Elsevier, Amsterdam.
Moubray, J. (1997), Reliability Centered Maintenance, 2nd ed., Industrial Press, New York, NY.
Murray, W. (2007), “Reliability centred knowledge”, available at: www.reliabilityweb.com
Nowlan, F., Heap, S. and Howard, F. (1978), Reliability Centered Maintenance, Dolby, Access
Press, New York, NY.
Pintelon, L., Nagarur, N. and Puyvelde, V.F. (1999), “Case study: RCM-yes, no or maybe?”,
Journal of Quality in Maintenance Engineering, Vol. 5 No. 3, pp. 182-92.
Place, S., Hall, P.L., Allsopp, K., Iriving, P.E. and Strutt, J.E. (2000), “Reliability prediction of
helicopter transmission systems with Bayesian updating by HUMS”, Final project,
EPSRCWU7204v/45, Cranfield University, Cranfield.
Ross, T.J. (1997), Fuzzy Logic with Engineering Applications, McGraw-Hill, New York, NY.
Sharma, R.K., Dinesh, K. and Pradeep, K. (2006), “Systematic failure mode effect analysis
(FMEA) using fuzzy linguistic modeling”, International Journal of Quality & Reliability
Management, Vol. 22 No. 9, pp. 986-1004.
Stewart, M.G. and Melchers, R.E. (1997), Probabilistic Risk Assessment of Engineering Systems,
Chapman & Hall, London, pp. 40-54.
Zadeh, L.A. (1975), “The concept of a linguistic variable and its application to approximate
reasoning pts. 1,2 and 3”, Information Science, Vol. 8, pp. 199-249, 301-57 and Vol. 9,
pp. 43-80.
Further reading
Brown, R.E. (2004), “Failure rate modelling using equipment inspection data”, IEEE
Transactions on Power Systems, Vol. 19 No. 2, pp. 782-7.
Hart, D.R. (2005), “Linking equipment reliability & optimum repair practices: a business-based
maintenance approach”, IEEE 0-7803-9107-I/05.
Ralph, A.E. (2002), “Data, data, oh where art thou, data?”, IEEE transactions on Reliability, Vol. 51
No. 3.
Zakrajesk, J.J., Townsend, D.O., Lewicki, D.G., Decker, H.J. and Handschuh, R.F. (1995),
“Transmission diagnostic research at NASA Lewis research centre”, paper presented at
I Mech E Conference on Gear Box Noise, Vibration and Diagnostics.
IJQRM
28,1
122
Downloaded by Indian Institute of Technology Kharagpur At 23:24 19 October 2015 (PT)
Appendix
Component Function
Failure
mode Effects on the system
Possible cause
of failure
(internal)
Possible predictive
monitoring Final maintenance action
Input and
output gear
pair
Transmit mechanical
torque
(F1)
Abnormal
sound
Unwanted mechanical
impacts and damage of
gearbox
Higher gear
backlash Body vibrations Adjust backlash
Input bearings
failed Bearing vibration Replace bearings
More axial
shift Bearing vibration Adjust axial play
(F2)
Abnormal
vibration
Broken Gear
Teeth
Ultrasonic crack detection/die
penetration test (off-line
monitoring)
Replace gearwheel followed
by visual inspection
(Endoscopy)
Less
lubrication Oil flow rate
Check and rectify
lubrication systems
Bearings
Provides friction less
support for rotation of
gear shaft
High
bearing
temp. (F3)
Lubricant damage. high
bearing wear Excessive load Bearing temp. Check bearings
High bearing
clearance Bearing vibration Replace bearings
Poor
lubrication Bearing temp.
Maintain lubrication
system
Poor heat
recovery Bearing temp.
Maintain – ventilation
system
Bearing
seizure
(F3a)
Secondary damage of
connected components No lubricant Bearing temp. Top up lubricant
Defective
bearing Bearing vibration Replace bearing
Indentation Body vibrations Improper gear mesh
Dislocation Housing vibration Check housings
Fatigue failure Bearing vibration Replace bearings
(continued)
Table AI.
Failure mode and effect
analysis of a steel mill
bevel gearbox (24.0 k
NM)
Prioritization of
maintenance
tasks
123
Downloaded by Indian Institute of Technology Kharagpur At 23:24 19 October 2015 (PT)
Component Function
Failure
mode Effects on the system
Possible cause
of failure
(internal)
Possible predictive
monitoring Final maintenance action
Bearing
housings
Offer stable support to
shaft with bearings
High
vibration
(F4)
Bearing damage.
increased stress on
rotating parts Looseness Housing vibration Tighten housing
Crack in the
housing Housing vibration Replace housing
Wear out Housing vibration Rework housings
Mis-alignment Bearing vibration Align
Eccentricity Bearing/body vibration
Align and check housings/
shaft bearings etc.
Input/
output
shaft Transmit torque
Shaft
shear (F5)
No mechanical input/
output Material defect No check Replace shaft
Excessive load Load measurement
Crack
propagation Ultrasonic crack/X-ray
Table AI.
IJQRM
28,1
124
Downloaded by Indian Institute of Technology Kharagpur At 23:24 19 October 2015 (PT)
Priority
Component Failure mode Possible predictive monitoring
Preventive maintenance
task
Without
POD
With
POD Final maintenance action
Input and output
gear pair
(F1) Abnormal
sound Body vibrations
Check gear teeth
condition 1 2 Adjust backlash
Bearing vibration Measure backlash Replace bearings
Bearing Vibration Measure axial play Adjust axial play
(F2) Abnormal
vibration
Ultrasonic crack detection/Die
penetration test (off-line
monitoring) Check gear teeth
Replace gearwheel followed by
visual inspection (Endoscopy)
Oil flow rate Flush oil system 2 1
Check and rectify lubrication
systems
Bearings
(F3) High
bearing temp. Bearing temp.
Measure bearing
Clearance visual check 4 3 Check bearings
Bearing vibration Replace bearings
Bearing temp. Maintain lubrication system
Bearing temp. Maintain ventilation system
(F3a) Bearing
seizure Bearing temp. Top up lubricant
Bearing vibration Replace bearing
Body vibrations Improper gear mesh
Housing vibration Check housings
Bearing vibration Replace bearings
Bearing housings
(F4) High
vibration Housing vibration Check fasteners 3 4 Tighten housing
Housing vibration Replace housing
Housing vibration Rework housings
Bearing vibration Align
Bearing/body vibration
Align and check housings/shaft
bearings etc.
Input/output
shaft (F5) Shaft shear No check Check shaft for cracks 5 2 Replace shaft
Load measurement
Ultrasonic crack/X-ray
Table AII.
Comparison of priority of
preventive tasks
Prioritization of
maintenance
tasks
125
Downloaded by Indian Institute of Technology Kharagpur At 23:24 19 October 2015 (PT)
About the authors
Edwin Vijay Kumar, Fellow of Indian Institute of Engineers (India), obtained his Bachelor’s
Degree in Electrical and Electronics Engineering in 1984 and works with an Integrated Steel
Plant, India. He has worked for almost 15 years in the area of Condition Monitoring and
Machinery Diagnostics. He is presently doing Research in Reliability Engineering at Indian
Institute of Technology, Kharagpur. Edwin Vijay Kumar is the corresponding author and can be
contacted at: Edwin_vijay@rediffmail.com
Sanjay Kumar Chaturvedi obtained his BE (Elect.) and ME (System Eng. & Operations
Research) degrees from IIT, Roorkee, India in 1988 and 1990, respectively. He completed his PhD
from Reliability Engineering Centre, IIT, Kharagpur, India in 2003. He is presently working as
Assistant Professor (Reliability Eng. Centre) at IIT, Kharagpur. His research interests are
Systems Reliability Evaluation, Optimization, Failure Data Analysis and Maintenance
Engineering.
IJQRM
28,1
126
To purchase reprints of this article please e-mail: reprints@emeraldinsight.com
Or visit our web site for further details: www.emeraldinsight.com/reprints
Downloaded by Indian Institute of Technology Kharagpur At 23:24 19 October 2015 (PT)
This article has been cited by:
1. Abdul Hameed, Faisal Khan, Salim Ahmed. 2015. A risk-based methodology to estimate shutdown
interval considering system availability. Process Safety Progress 34:10.1002/prs.v34.3, 267-279. [CrossRef]
2. Pritibhushan Sinha. 2015. Towards higher maintenance effectiveness. International Journal of Quality &
Reliability Management 32:7, 754-762. [Abstract] [Full Text] [PDF]
3. Ruy Victor B. de Souza, Luiz Cesar R. Carpinetti. 2014. A FMEA-based approach to prioritize waste
reduction in lean implementation. International Journal of Quality & Reliability Management 31:4,
346-366. [Abstract] [Full Text] [PDF]
Downloaded by Indian Institute of Technology Kharagpur At 23:24 19 October 2015 (PT)