A Combined Goal Programming AHP Approach To Maintenance Selection Problem

You might also like

You are on page 1of 10

Reliability Engineering and System Safety 91 (2006) 839–848

www.elsevier.com/locate/ress

A combined goal programming—AHP approach


to maintenance selection problem
Massimo Bertolini a,*, Maurizio Bevilacqua b
a
Dipartimento di Ingegneria Industriale, Università degli Studi di Parma, Viale delle Scienze, 181/A, 43100 Parma, Italy
b
Dipartimento di Ingegneria delle Costruzioni Meccaniche, Nucleari, Aeronautiche e di Metallurgia, sede di Forlı̀,
Università degli Studi di Bologna, Via Fontanelle 40, 47100 Forlı̀, Italy
Received 14 October 2004; accepted 18 August 2005
Available online 10 October 2005

Abstract
This paper presents a ‘Lexicographic’ Goal Programming (LGP) approach to define the best strategies for the maintenance of critical
centrifugal pumps in an oil refinery.
For each pump failure mode, the model allows to take into account the maintenance policy burden in terms of inspection or repair and in terms
of the manpower involved, linking them to efficiency-risk aspects quantified as in FMECA methodology through the use of the classic parameters
occurrence (O), severity (S) and detectability (D), evaluated through an adequate application of the Analytic Hierarchy Process (AHP) technique.
An extended presentation of the data and results of the case analysed is proposed in order to show the characteristics and performance of this
approach.
q 2005 Elsevier Ltd. All rights reserved.

Keywords: Maintenance strategy selection; Goal programming; Analytic hierarchy process

1. Introduction intervention on the system and safety hazards. For that reason,
an appropriate maintenance policy strategy is necessary in
Proper maintenance of plant equipment can significantly order to replace the deteriorated system before failure [2]. The
reduce the overall operating cost, while boosting the managers must select the best maintenance policy for each
productivity of the plant. Management personnel often piece of equipment or system out of a set of possible
consider plant maintenance an expense, yet a more positive alternatives. In particular, the development of maintenance
approach is to view maintenance work as a profit centre. In strategies (i.e. a suitable combine between corrective,
consideration of this new perspective, the requirements for preventive and predictive maintenance policies) must take
maintenance management have change drastically from the old into account that resources are limited and, as a result, that
concept of ‘fix-it-when-broken’ to a more complex approach, maintenance activities should be imperfect [3].
which entails adopting a maintenance strategy for a more Several methods have been proposed in the literature for
integrated approach and alignment. Furthermore, the high level planning maintenance activities of industrial plants. Among
of complexity of today’s industrial plans requires an elevated these, the Reliability Centred Maintenance (RCM) approach
level of availability and reliability of such systems. The [4] offers a systematic and efficient Decision Support System
development of new technologies and managerial practices (DSS) tool to the optimisation of plant and equipment
means that maintenance staff must be endowed with growing maintenance. In particular, RCM approach is designed to
technical and managerial skills [1]. minimise maintenance costs by balancing the costs of different
The deterioration and failure of systems might incur high maintenance strategy taking into account the loss of potential
costs, due to production losses and delays, unplanned life [5]. Recent DSS study [6] highlights the benefits of an
Integrated Multi Criteria Decision Making approach when
tackling Reliability, Availability, Maintainability and Safety
* Corresponding author. Tel.: C39 0521 905861; fax: C39 0521 905705. (RAMS) problems in complex process plants like a nuclear
E-mail address: massimo.bertolini@unipr.it (M. Bertolini). power plant, from a technical specification and maintenance
0951-8320/$ - see front matter q 2005 Elsevier Ltd. All rights reserved. point of view. The use of the proposed methodology that
doi:10.1016/j.ress.2005.08.006 balances RAMS and costs (RAMSCC) allows to
840 M. Bertolini, M. Bevilacqua / Reliability Engineering and System Safety 91 (2006) 839–848

simultaneously satisfy the relevant criteria (i.e. reliability, criticality for each of the above classes of consequences.
availability, maintainability, safety, costs, human resources) Several RCM concepts applications can be found in literature:
with respect of the constraints conditions. Another interesting Deshpande and Modak describe a systematic RCM application
DSS maintenance analysis is described in [7]. Based on the to a steel melting shop of a medium-scale industry [12], taking
integration of operational and maintenance linked aspects in also into account safety consideration to develop consistent
chemical process plants, the authors develop a framework that strategies to assess the maintenance problems in modern
allows to identify process and availability requirements at the manufacturing industry [13] and analysing the problems that
design stage. This allow to clearly defining the optimal trade arise in applying RCM technique in western and eastern firms
off between investment and operational cost from an [14]. Carratero et al. [15] describe the application of RCM to
availability point of view. The study of Cowing et al. [8] large system like a railway network, stressing that an
analyses the long-term evolution of trade off between safety interdisciplinary approach can provide optimal results for an
and productivity taking into account the influence of efficient and effective maintenance implementation.
interruption of operation due to scheduled and on demand To improve RCM implementations, scheduling mainten-
maintenance, providing a model that allows to optimally ance activities, Vatn [16] suggested combining decision theory
manage the different operation strategies in order to maximise models and influencing diagrams. The framework proposed by
long-term performance. the author allows to perform quantitative analyses of
This paper represents a new contribution in the field of maintenance strategy incorporate preferences and value trade-
Decision Support System (DSS) tool for maintenance policies offs explicitly in the analyses. Another discussion on
selection problem. In particular, a Goal Programming approach optimization models applied to maintenance decision making
is proposed to minimise maintenance costs, solving a can be found in [17], where the authors discuss the application
maintenance strategy selection problem concerning the various of operational and strategic decision support systems to
centrifugal pumps present in an oil refinery plant. The use of maintenance, stressing the necessity of new information
Goal Programming allows considering multiple criteria to technology tool to take advantage of the potential of
measure performance, multi object/goals and constraints. The optimization models.
use of the pre-emptive goals technique especially allows the Waeyenbergh and Pintelon [18] developed and
analyst to assign different priorities to the goals considered, implemented a 7-step modular framework to provide a support
looking for a solution that first of all meets the most important when deciding which maintenance policy to use. This
of these priorities. framework was developed on the basis of the well-known
maintenance concepts present in literature (i.e. RCM, TPM and
2. Maintenance strategy selection problem BCM [19]).
Others Multi Criteria Decision (MCD) approaches have
Estimating the best set of maintenance policies for the been suggested in the recent years. In particular, Almeida and
different failure modes is a hard and complex task. This Bohoris [20] consider the application of decision-making
selection requires the knowledge of various factors such as theory to maintenance paying special attention to the multi-
safety aspects, environmental problems, costs and budget attribute utility theory. Triantaphyllou et al. [21] suggest the
constraints, manpower utilization, MTBF and MTTR for each use of the Analytical Hierarchy Process (AHP) considering
piece of equipment, etc. The management of the large number only four maintenance criteria: cost, reparability, reliability
of tangible and intangible attributes that must be taken into and availability. Bevilacqua and Braglia [22] presented an
account represents the main complexity of the problem. application of the AHP technique for maintenance strategy
In spite of the great importance of the problem, there are not selection in an Italian oil refinery processing plant, combining
many studies that deal with the analysis and development of many features which are important in the selection of the
maintenance policy selection. Some Decision Support System maintenance policy: economic factors, applicability and costs,
(DSS) approaches are proposed in literature as an aid to the safety, etc.
solution of this problem. As mentioned above, Reliability- In order to make the cost control more effective, an
Centred Maintenance (RCM) [4] is one of the best known and integrated AHP and GP approach is suggested in this paper: the
most used devices to preserve the operational efficiency of use of a combined model allows to investigate the maintenance
production tools. RCM operates by balancing the high selection problem in detail, taking into account the resource
corrective maintenance costs with the cost of programmed burden and providing the analyst with a tool to assess the
(preventive or predictive) policies, taking into account the priority level of the different maintenance alternatives.
potential shortening of the ‘useful life’ of the item considered.
Its principles are widely described in [9–11] and its application 3. Goal programming technique
ranges from military, nuclear power plants and electric power
generation, to many other industrial sectors. Rausand [4] Goal programming is a well-known modification and
describes the guidelines for a practical application of RCM extension of linear programming, developed in the early
concepts identifying four different consequences of a failure, 1960s owing to the study of Charnes and Cooper [23]. Linear
i.e. safety of personnel, environmental impact, production programming deals with only one single objective to be
availability and cost of material loss, defining four levels of minimized or maximized, and subject to some constraint; it,
M. Bertolini, M. Bevilacqua / Reliability Engineering and System Safety 91 (2006) 839–848 841

therefore, has limitations in solving a problem with multiple Taking into account m objectives we have:
objectives. Goal programming, instead, can be used as an X
m
effective approach to handle a decision concerning multiple min Z Z Pj ðdjK; dC
j Þ (1)
and conflicting goals. Also, the objective function of a goal- jZ1
programming model may consist in non-homogeneous units of
The deviations are mutually exclusive and this lead to the
measure.
condition expressed in Eq. (2):
The goal-programming approach is extensively applied in
decision analysis in operations research, such as production djKdC
j Z0 (2)
planning, financial decisions [24], marketing decisions,
The Pj factors reflect the problem hierarchy: P1 represent
corporate planning, academic planning, and decision in
the highest level, P2 the second priority level, and so on.
government [25,26]. In particular, it has been successfully
The objective function is subject to m constraint equations,
used to solve several Multi Criteria Decision (MCD)
as shown in Eq. (3)
problems, such as the design of a quality control procedure
" #
in service organizations [27]; the selection of the optimal set Xn
of service quality control instruments [28]; regarding ðaji xi Þ C djKKdCj Z Bj ; with j Z 1; 2; .; m (3)
information system selection [29] and the identification and iZ1

development in the mathematical model for information where


system project selection in health service institutions; and,
finally, to help the facility planning authorities to formulate Bj represents the objective target of the jth resource;
viable location strategies in the volatile and complex global aji is the usage of jth resource of every possible
decision environment [30]. alternative ith decision.
The different goal programming models available to assess
MCD problems include the non-linear and linear GP with
Archimedean weights (i.e. weighted GP), the Interactive
Weighted Tchebycheff Procedure (IWT), the MINMAX 4. The AHP-GP model for maintenance policy selection
(Chebyshev) GP, the Reference Point Method (RPM), the
The combined AHP-GP model embodies AHP results in the
Compromise Programming (CP), and the Lexicographic
GP model. In particular, in the model described here the AHP
Linear GP.
analysis provides the priority vector of the possible mainten-
Lexicographic goal programming is actually one of the
ance policies (corrective, preventive and predictive) for each
most significant devices in tackling MCD problems: the
failure type revealed. The use of AHP allows to define a three-
different goals can be ranked according to different priority
level hierarchical structure: the top level represents the goal of
levels that reflect the target allocated to them by the decision
the analysis (in this case the maintenance policy definition), the
maker.
second level is relative to the relevant criteria used (occurrence,
The lexicographic approach defines different priority levels
severity and detectability), the third one defines the possible
Pj for the goals of the analysis. The different priority levels alternatives.
reflect the hierarchical relationship between the targets in the AHP is a decision-making procedure originally developed
objective function where they are arranged in order of by Saaty in the 1970s and described in Saaty [32]. Some
decreasing priority (P1OP2O/OPm). criticism has been expressed by decision analysts on the use of
In order to identify the solution to the problem, the highest AHP for multi criteria decision making, mainly based on the
priority goals and constraints are considered first; if more than lack of a strong normative foundation and on a possible
one solution is found in the first step, another goal ambiguity of the questions the decisions maker must answer
programming problem is formulated which takes into account (see, for example, [33,34]). The replies of Saaty and Vargas to
the second priority level targets. The procedure is repeated those critics can be found, as en example, in references
until a unique solution is found, gradually considering [35–38]. Recently, Bana and Vansnick [39] motivate the new
decreasing priority levels. critic on AHP technique based on the missing respect of a
The lexicographic optimisation can then avoid the estimate measurement condition. Although several researches express
of the different deviation weights, but the results of the analysis discordant opinion on the use of AHP, the technique is still
may be biased by the analyst’s personal opinion [31]. widely used in practice.
In this paper, the LGP model is applied defining a binary Once the AHP qualitative structure had been defined,
structural variable (zero-one programming), and the objective the data for the pairwise comparison necessary to obtain the
function reported in Eq. (1) shows that the goal of the problem maintenance policy scores were collected by interviewing the
consists in the minimization of the unwanted deviations from oil refinery maintenance and production management staff. The
the target. AHP analysis outcomes used to develop the AHP-GP model
In Eq. (1), dKj and dj , respectively, represent the negative
C
are the global priority (SCOREAHP,i) of the different ith
and positive deviation from the value, either desired or alternatives (the possible maintenance policies for every failure
constrained, of the jth objective. type), the local priority SCOREk,i of the ith alternative with
842 M. Bertolini, M. Bevilacqua / Reliability Engineering and System Safety 91 (2006) 839–848

respect to each criterion, and the weights (wk) of the kth criteria 1. cost minimization;
(occurrence, severity and detectability) present at the second 2. manpower usage minimization;
level of the hierarchical structure developed. The final outcome 3. AHP score maximization;
is a vector, normalised to the unity, that allows to identify the 4. local score maximization for each criterion (occurrence,
better alternative with respect to the target. AHP technique also severity and detectability).
allows the analyst to evaluate the correctness and the
consistency of the given pairwise comparisons, by means of Once the cost and manpower usage targets have been
an inconsistency ratio (IR) [32]. The judgments can be defined, it is possible to formulate the complete AHP-GP
considered acceptable if and only if IR%0.1. model reported in Eq. (5) (for a complete list of notation used
The objective function can thus be formulated as in Eq. (4): in this paper, the reader may refer to the Appendix A):
X X
Pk ðwk dkK; wk dC Pj ðdjK; dC min Z Z P1 ðdC
C Þ C P2 ðdMT Þ C P3 ðdSCORE;AHP Þ C P4 ðwO dO
C K K
min Z Z lk Þ C j Þ (4)
k j
S C w D dD Þ
C wS dK K

The goal is the minimization of the unwanted deviations,


also by taking into account the AHP scores. The subject to

CCORR xCORR C CPREV xPREV C CCOND xCOND C dK


C KdC Z TC
C

MTCORR xCORR C MTPREV xPREV C MTCOND xCOND C dK


MT KdMT Z TMT
C

SCOREAHP;CORR xCORR C SCOREAHP;PREV xPREV C SCOREAHP;COND xCOND C dK


SCORE;AHP KdSCORE;AHP Z 1
C
(5)
SCOREO;CORR xCORR C SCOREO;PREV xPREV C SCOREO;COND xCOND C dK
O KdO Z TO;SCORE
C

SCORES;CORR xCORR C SCORES;PREV xPREV C SCORES;COND xCOND C dK


S KdS Z TS;SCORE
C

SCORED;CORR xCORR C SCORED;PREV xPREV C SCORED;COND xCOND C dK


D KdD Z TD;SCORE
C

lexicographic structure of the goal programming model


considers the (wk) as sub-weights since they are weighted In the model above, the two first constraint equations are
through the priority scores Pk. relative to the maintenance cost and manpower usage targets,
The problem solution, obtained using LINDO software, more specifically:
allows the analyst to point out the relative importance of the
different maintenance policies taking however into account the (i) CCORR, C PREV , C COND and MT CORR , MTPREV ,
limitation of the resources available. MTCOND represent the cost and maintenance time for
The maintenance policy definition was obtained through the corrective, preventive and predictive policies;
AHP (see Fig. 1) on the basis of the classical FMECA (ii) TC and TMT indicate the availability of the maintenance
parameters occurrence, severity and detectability defining the resources;
C ; dC and dMT ; dMT define the negative and positive
(iii) dK C K C
criteria’s respective weights wO, wS, wD.
deviations for the maintenance cost and the mainten-
The structural binary variables xCORR, xPREV, xCOND
ance time, respectively.
represent corrective, preventive and predictive maintenance
policies, and the objectives, listed as decreasing priority values, The above equations are linked to the further model
have been identified as: objectives ‘AHP score maximization’ and ‘local score
maximization’: SCOREAHP,i is the score obtained by the
ith maintenance policy alternatives through AHP analysis,
and SCOREk,i is the local score of the ith alternative with
respect to the kth criteria under examination (occurrence,
severity and detectability). The Tk,SCORE values represent
the targets defined for the constraint equations linked to
the local score maximization score. The numerical values
of the Tk,SCORE, have been chosen, according to Badri [28],
as the sum of the two highest SCOREk,i values. The
deviations from the target are referred to as dK k ; dk for
C

each criterion.
The objective function aims at minimizing the sum of the
deviations associated to each specified goal, taking into
Fig. 1. AHP hierarchy for maintenance policy selection. account only the unwanted deviations.
M. Bertolini, M. Bevilacqua / Reliability Engineering and System Safety 91 (2006) 839–848 843

Table 1
Pump operating parameters

Item Production plant Fluid Fluid tempera- Flow rate (m3/h) Head (m) Density (kg/m3) NPSH (m)
ture (8C)
Pu1 Topping Crude oil 18 530 237 870 6
Pu2 Topping Resin 164 55 73 661 2
Pu3 Topping Bitumen 360 231 173 801 4
Pu4 Topping Gas oil 257 123 108 668 3.8
Pu5 Topping Gas oil 187 63 117 670 2.3
Pu6 Topping Gas oil 326 42 92 663 2
Pu7 Topping Gas oil 188 459 192 764 4.3
Pu8 Topping Gasoline 38 151 205 694 2.5
Pu9 Unifining Hydrocarbon 239 687 146 468 2.9
Pu10 Unifining Hydrocarbon 40 172 67 669 3.6

5. The case study Failure and maintenance data relative to a 10-year period
were analysed, gathering information such as failed item,
The AHP-GP model described above has been applied to failure cause and effect, MTBF, MTTR, repair cost. The
identify the optimal maintenance policy for a set of centrifugal analysis allowed to highlight the 10 most critical pumps,
pumps operating in the process and service plants of an Italian which, in agreement with the oil refinery maintenance
oil refinery. The data necessary to the analysis were collected management, were chosen to formed the set for the application
through specific queries to the oil refinery Computerized of the AHP-GP model. Such pumps will be referred to in the
Maintenance Management System (CMMS). following as Pu1, Pu2, Pu3, Pu4, Pu5, Pu6, Pu7, Pu8, Pu9,

Table 2
Pump failure types

Item Failure
Component Effect Cause
Pu1 Coupling Vibration Misalignment
Double seal Damage Foreign body/normal operation
Pu2 Coupling Vibration Misalignment
Double seal Damage Foreign body/normal operation
Pu3 Inlet gate Fluid loss Breakdown
Coupling Vibration Misalignment
Double seal Damage Foreign body/normal operation
Pu4 Inlet gate Fluid loss Breakdown
Pump body Fissure Inhibited dilatation
Strain Deterioration Foreign body/normal operation
Flow meter Deterioration Foreign body/normal operation
Double seal Damage Foreign body/normal operation
Inlet pipe Breakdown Aggressive agent/uncontrolled dilatation
Pu5 Pump body Fissure Inhibited dilatation
Double seal Damage Foreign body/normal operation
Servo valve Malfunction Wear/normal operation
Pu6 Single seal Damage Foreign body/normal operation
Double seal Damage Foreign body/normal operation
Pu7 Bearing Damage Bad lubrication
Strain Blockage Foreign body/normal operation
Double seal Damage Foreign body/normal operation
Pu8 Strain Blockage Foreign body/normal operation
Single seal Damage Foreign body/normal operation
Pu9 Pump shaft Flexure Thermal shock/bearing breakdown
Brass Wear Bad lubrication/normal operation
Pump body Erosion Foreign body/normal operation
Double seal Damage Foreign body/normal operation
Pu10 Brass Wear Bad lubrication/normal operation
Pump body Erosion Foreign body/normal operation
Bearing Damage Bad lubrication
Pump impeller Damage Wrong assembly/foreign body
Single seal Damage Foreign body/normal operation
844 M. Bertolini, M. Bevilacqua / Reliability Engineering and System Safety 91 (2006) 839–848

Table 3
Possible predictive and preventive action for the failure type

Failure type Preventive maintenance Predictive maintenance action Notes


action
Coupling misalignment Alignment control1 Not possible 1
Through stroboscope
Seal damage Visual inspection2 Flux fluid analysis 2
For single seals
Inlet gate breakdown Not possible Strain analysis3 3
Through strain gage and pipe tracing for a proper
heating sequence
4
Breakdown due to pump Not possible Integrity control with dye penetrant Failure due to an inlet gate strain
body fissure4
Strain blockage Functionality control5 Functionality control6 5
Through the use of strains with visual inspection
capability; 6Through a differential-pressure gage
with alert to control the strain pressure drop
Flow meter wrong Functionality control7 Not possible 7
Periodic control of instrument tuning and fluid
measurement passing with a flow indicator
Pipeline breakdown8 Not possible Thickness control9 8
Failure due to aggressive agent erosion and
uncontrolled strain; 9Through ultrasonic thickness
gage
Servo valve breakdown Periodic substitution at a Not possible
fixed number of
operating cycles
10
Bearing breakdown Not possible Temperature and lubricating oil analysis through Assess and report on temperature, oil condition,
temperature probe and oil monitoring systems, speed, bearing condition, shaft alignment, noise,
vibration analysis through accelerometer10 vibration and cavitation
11
Shaft flexure Not possible Vibration and noisiness analysis and electric motor An increase in electric motor absorption can
absorption control11 indicate the presence of pump cavitation
Brass wear Substitution at a fix Not possible
operating period
Brass high vibration level Periodic inspection of the Not possible
lubrication system
12
Pump body erosion Protective elements Not possible Generally performed during plant turnaround
(liner) insertion to limit
pump body erosion12
Impeller damage13,14 Not possible Vibration analysis through accelerometer14 13
Failure due to a bad strain operation;
14
Phenomenon due to wrong assembly

Pu10; their relevant operating parameters are reported in The corrective maintenance cost, distinguished for the
Table 1 and the failure type (classified for item, effect and several failure types, was calculated as the mean of the data
cause) in Table 2. from the past 10 years, updating the values to the last period of
The aim of the analysis is to identify the optimal operation. The maintenance cost comprises either item repair
maintenance policy for each failure type, taking into account or substitution fee and the manpower expense, evaluated on the
the feasibility of the different alternatives. Table 3 reports a basis of the labour-hours needed.
brief description of the possible predictive and preventive API oil refinery outsourcers the management of the
action for the critical items identified per failure type. inspection policy to a third party firm; the maintenance
The pairwise comparison data needed to assess the priority
of the three criteria used (i.e. occurrence, severity and Table 4
Priority vector for criteria used
detectability) were collected by interviewing the maintenance
engineering technical and management staff of API oil Criteria Weight (wk)
refinery. The resulting priority vector (wk) is shown in Severity wS Z 0:724
Table 4. The overall inconsistency ratio (IR) results to be Occurrence wO Z 0:193
Detectability wD Z 0:083
0.044 (%0.1), thus ensuring the consistency of the pairwise
comparison. Also the partial IR values, not reported for Inconsistency ratio (IR)Z0.044.
brevity, are all lower than 0.1.
The most relevant criterion appears to be severity, directly Table 5
Example of priority index for the three maintenance policies considered
linked to safety and production aspects. Table 5 reports, as an
example, the priority index for the three maintenance policies Failure type Corrective Preventive Predictive main-
here considered relating to the failures ‘single seal damage’ maintenance maintenance tenance priority
priority priority
and ‘blocked strain’.
To apply the AHP-GP model, maintenance time and Single seal 0.085 0.275 0.640
damage
cost data were collected by querying the oil refinery
Blocked strain 0.126 0.405 0.469
CMMS.
M. Bertolini, M. Bevilacqua / Reliability Engineering and System Safety 91 (2006) 839–848 845

Table 6
The maintenance cost and time data

Items Failure type Corrective Preventive Predictive Minutes for Minutes for Minutes for TC TMT
cost cost cost corrective preventive predictive
mainten- mainten- mainten-
ance ance ance
Pu1 Coupling Vibration V1184 V875 – 360 252 – 1033 1200
Double seal Damage V1697 – V1141 720 – 612 2840 1200
Pu2 Coupling Vibration V514 V504 – 240 168 204 1033 1200
Double seal Damage V1453 – V1093 360 252 306 2840 1200
Pu3 Inlet gate Fluid loss V1552 – V1112 360 252 306 1136 1200
Coupling Vibration V254 V348 – 240 168 204 620 1200
Double seal Damage V8369 – V4769 900 630 765 5939 1200
Pu4 Inlet gate Fluid loss V548 – V561 360 252 306 1050 1200
Pump body Fissure V1585 – V1236 1080 756 918 1446 900
Strain Deterio- V943 V762 V1081 240 168 204 774 900
ration
Flow meter Deterio- V289 V339 – 360 252 306 413 900
ration
Double seal Damage V4716 – V2887 360 252 306 3357 1200
Inlet pipe Breakdown V161 – V418 60 42 51 516 900
Pu5 Pump Fissure V903 – V861 1080 756 918 1446 1200
Double seal Damage V4079 – V2281 1440 1008 1224 2840 1200
Servo valve Malfunc- V222 V298 – 360 252 306 774 1080
tion
Pu6 Single seal Damage V431 V424 V496 360 252 306 1033 200
Double seal Damage V2200 V1392 – 720 504 612 2840 1200
Pu7 Bearing Damage V819 – V351 1240 868 1054 1420 1200
Strain Blockage V325 V391 V742 240 168 204 774 300
Double seal Damage V3398 – V2077 720 504 612 2840 1200
Pu8 Strain Blockage V483 V486 V828 240 168 204 774 900
Single seal Damage V6235 V3721 V3553 1080 756 918 4131 1200
Pu9 Pump shaft Flexure V1889 – V973 1440 1008 1224 1239 900
Brass Wear V454 V407 – 480 336 408 930 900
Pump body Erosion V1424 V1020 – 360 252 306 1213 900
Double seal Damage V5931 V3631 V3471 720 504 612 4131 800
Pu10 Brass Wear V384 V365 – 480 336 408 930 900
Pump body Erosion V231 V304 – 360 252 306 516 900
Bearing Damage V2924 – V1735 480 336 408 2840 1200
Pump Damage V2856 – V1942 180 126 153 2840 900
impeller
Single seal Damage V6479 V4053 V3857 360 252 306 4131 1200

cost was then evaluated on the basis of a fixed operating time equal to the mean number of failures increased by one and the
item substitution or inspection outcomes (preventive policy), maintenance time was determined as a fixed percentage aZ
or on the basis of the real time analysis of the controlled 70% of the corresponding corrective maintenance time, as
items operating parameters (predictive policy). The values shown in Eq. (7):
above were added to the manpower cost, estimated through
the mean preventive and predictive inspection times. MTPREV Z TPREV NPREV ; TPREV Z TFAILURE a;
The estimate of the preventive and predictive maintenance (7)
NPREV Z NFAILURE C 1
times for the failure types considered for the observed pumps
was carried out as in the following: for a given pump and for For a predictive maintenance we suppose that:
each cause of failure we have:
MTCOND Z TCOND NCOND ; TCOND Z TFAILURE b;
MTCORR Z TFAILURE NFAILURE (6) (8)
NCOND Z NFAILURE
where MTCORR, TFAILURE, NFAILURE are the overall corrective
maintenance time, the single item repair time and the expected where b is higher than a (bZ85%) in consideration that the
number of failures of the item during the observation period. mean time for predictive maintenance is considered comprised
As far as preventive maintenance policy is considered, the between the corrective and preventive maintenance time
number of programmed interventions was hypothesized as values.
846 M. Bertolini, M. Bevilacqua / Reliability Engineering and System Safety 91 (2006) 839–848

The proposed evaluation method appears to be rather rough. The analysis of the results points out that in most cases the
However, in consideration of the kind and amount of data goals of the model are reached by choosing the predictive
collected and the problem dimension, a more accurate maintenance policy.
statistical analysis (i.e. failure rate trend vs. item operating For 21 failure types, the optimal maintenance is the
time or optimal programmed maintenance mathematical predictive type, in 10 situations it is better to adopt a preventive
modelling) reveals itself as not viable. maintenance policy, and in one case it is convenient to combine
The maintenance cost and time data for the three different preventive and predictive policies. In no cases the model
policies are summarized in Table 6, where TC and TMT are the recommended the corrective policy.
budget and maintenance time availability, respectively.
The model implementation also requires to define the upper 5.1. Result discussion
bound values of the resources involved in the maintenance
task. These bounds have been fixed to a level not higher than The AHP-GP model proposed highlights that in the situation
the mean values recorded during the operating periods studied predictive maintenance obtains the best compromise
preceding relevant revamping of the controlled plant item. In between resource usage and reduction of failure effects. This
the case study here described such a period was always less fact is non-surprising since this policy allows failure
prediction, thus providing the maintenance manager with
than 5 years.
useful information to limit the negative failure aspects in terms
The AHP-GP model described in the above was solved
of both safety and cost.
using LINDO software, finding out the optimal maintenance
The model, when applied to the situation analysed, never
policy for every item failure type. The results of the analysis
suggested the adoption of corrective maintenance policy. Such
are summarized in Table 7.
a policy is characterized by a low investment for the tools
requested to perform the maintenance, but does not allow any
kind of failure prevention, with possible critical consequences
Table 7 in cost and safety for both people and production plants,
The results of the analysis especially firms in the petrochemical industry.
Item Failure Results The model outcomes were submitted to the maintenance
Component Effect Optimal maintenance mix staff of the oil refinery, whose management, on the basis of the
model results, decided to introduce a more accurate inspection
Pu1 Coupling Vibration Preventive
Double seal Damage Preventive plan for the critical failures of the items analyzed, thus
Pu2 Coupling Vibration Preventive reducing the possibility of catastrophic failures and allowing
Double seal Damage Predictive the maintenance function to better schedule activities
Pu3 Inlet gate Fluid loss Predictive improving cost control and operational efficiency.
Coupling Vibration Preventive
Double seal Damage Predictive 6. Conclusions
Pu4 Inlet gate Fluid loss Predictive
Pump body Fissure Predictive
This paper proposes a goal programming approach to the
Pu4 Strain Deterioration Preventive
Flow meter Deterioration Preventive selection of maintenance strategies for the centrifugal pumps of
Double seal Damage Predictive an oil refinery plants. This approach can simultaneously handle
Inlet pipe Breakdown Predictive the multiple and conflicting goals characteristic of decision
Pu5 Pump Fissure Predictive problems such as quality control system selection, facility
Double seal Damage Predictive location allocation problems, information system project
Servo valve Malfunction Preventive
evaluation, fund allocation. The combined AHP-GP model
Pu6 Single seal Damage PreventiveC Predictive
Double seal Damage Predictive
was applied in two subsequent stages: the first part of the
Pu7 Bearing Damage Predictive analysis provided the priority levels for the different
Strain Blockage Predictive maintenance policies with respect to the classical FMECA
Double seal Damage Predictive criteria (occurrence, severity and delectability), the second
Pu8 Strain Blockage Preventive step, with the formulation of the Goal Programming model, has
Single seal Damage Predictive led to the identification of the best set of maintenance type for
Pu9 Pump shaft Flexure Predictive the equipment failure modes considered.
Brass Wear Preventive
Pump body Erosion Preventive
The decision model proposed compares three alternative
Double seal Damage Predictive maintenance strategies (corrective, preventive and predictive),
Pu10 Brass Wear Predictive taking into account (i) budget and (ii) amount of hours of
Pump body Erosion Predictive manpower labour constraints.
Bearing Damage Predictive The application of the GP technique combined with AHP
Pump impel- Damage Predictive
methodology proved to be a flexible tool to optimally allocate
ler
Single seal Damage Predictive the resource to the different maintenance strategies, a feature
that is particularly important in situations where the decision
M. Bertolini, M. Bevilacqua / Reliability Engineering and System Safety 91 (2006) 839–848 847

maker can choose between different objectives subject to deviations for the maintenance cost, dMTK
; dMTC

several constraint conditions. are the negative and positive deviations for the
The method here presented can provide a framework to maintenance time);
guide future investigations. In particular, in future works other MTCORR the overall corrective maintenance time;
kinds of goals and/or constraints could be potentially MTPREV the overall preventive maintenance time;
considered and added to the original model proposed. MTCOND the overall predictive maintenance time;
TFAILURE the single item repair time (TPREV is the item
Acknowledgements repair time in preventive policy and TCOND is the
item repair time in predictive policy);
The authors are indebted to the Referee for their NFAILURE the expected number of failures of the item
constructive comments which enabled the improvement of during the observation period (NPREV is the
the quality of the present study. number of programmed interventions in pre-
ventive maintenance, NCOND is the number of
Appendix A. Notation used in the paper programmed interventions in predictive
maintenance);
djK the negative deviation from the value desired
or constrained, of the jth objective;
djC the positive deviation from the value desired References
or constrained, of the jth objective;
Pj factors reflect the problem hierarchy (i.e. P1 [1] Smidt-Destombes KS, Heijden MC, Harten A. On the availability of a
represent the highest level, P2 the second k-out-of-N system given limited spares and repair capacity under a
priority level, and so on); condition based maintenance strategy. Reliab Eng Syst Safe 2004;83(3):
287–300.
Bj the objective target of the jth resource; [2] Hontelez JAM, Burger HH, Wijnmalen DJD. Optimum condition-based
aji the usage of jth resource of every possible maintenance policies for deteriorating systems with partial information.
alternative ith decision; Reliab Eng Syst Safe 1996;51(3):267–74.
xi the alternative ith decision (i.e. xCORR is the [3] Yuo-Tern Tsai, Kuo-Shong Wang, Lin-Chang Tsai. A study of
corrective maintenance policy); availability-centered preventive maintenance for multi-component
systems. Reliab Eng Syst Safe 2004;84(3):261–70.
SCOREAHP,i the global priority of the different ith alternatives;
[4] Rausand M. Reliability centred maintenance. Reliab Eng Syst Safe 1998;
SCOREk,i the local priority of the ith alternative with 60(2):121–32.
respect to each criterion; [5] Crockera J, Kumarb UD. Age-related maintenance versus reliability
wk the weights of the kth criteria (i.e. wO is the centred maintenance: a case study on aero-engines. Reliab Eng Syst Safe
weights of occurrence criteria); 2000;67(2):113–8.
[6] Martorell S, Villanueva JF, Carlos S, Nebot Y, Sanchez A, Pitarch JL,
CCORR the cost for the corrective policy;
et al. RAMSCC informed decision-making with application to multi-
CPREV the cost for the preventive policy; objective optimization of technical specifications and maintenance using
CCOND the cost for the predictive policy; genetic algorithms. Reliab Eng Syst Safe 2005;87(1):65–75.
MTCORR the maintenance time for the corrective policy; [7] Goel HD, Grievink J, Herder PM, Weijnen MPC. Integrating reliability
MTPREV the maintenance time for the preventive policy; optimization into chemical process synthesis. Reliab Eng Syst Safe 2002;
MTCOND the maintenance time for the predictive policy; 78(3):247–58.
[8] Cowing MM, Elisabeth Paté-Cornell M, Glynn PW. Dynamic modeling
TC the availability of the maintenance budget of the tradeoff between productivity and safety in critical engineering
resources; systems. Reliab Eng Syst Safe 2004;86(3):269–84.
PMT the availability of the maintenance time [9] Moubray J. Reliability centred maintenance. Oxford: Butter Worth-
resources; Heinmann Ltd; 1991.
SCOREAHP,i the score obtained by the ith maintenance policy [10] Smith AM. Reliability centred maintenance. New York, NY: McGraw-
Hill; 1993.
alternatives through AHP analysis (i.e. SCOR- [11] Anderson AT, Neril L. Reliability-centred maintenance management and
EAHP,CORR is the score obtained by the corrective engineering methods. Amsterdam: Elsevier; 1990.
maintenance policy through AHP analysis); [12] Deshpande VS, Modak JP. Application of RCM to a medium scale
SCOREk,i the local score of the ith alternative with respect industry. Reliab Eng Syst Safe 2002;77(1):31–43.
to the kth criteria under examination (i.e. [13] Deshpande VS, Modak JP. Application of RCM for safety considerations
in a steel plant. Reliab Eng Syst Safe 2002;78(3):325–34.
SCOREO,CORR is the local score of occurrence
[14] Deshpande VS, Modak JP. Maintenance strategy for tilting table of rolling
alternative with respect to the corrective mill based on reliability considerations. Reliab Eng Syst Safe, 80(1):1–18.
maintenance policy); [15] Carretero J, Perez JM, Garcia-Carballeira F, Calderon A, Fernandez J,
Tk,SCORE the targets defined for the constraint equations Garcia JD, et al. Applying RCM in large scale systems: a case study with
linked to the local score maximization score (i.e. railway networks. Reliab Eng Syst Safe 2003;82(3):257–73.
[16] Vatn J. Maintenance optimisation from a decision theoretical point of
TO,SCORE is the targets defined for the occur-
view. Reliab Eng Syst Safe 1997;58(2):119–26.
rence); [17] Dekker R, Scart PA. On the impact of optimisation models in
dkK; dkC the deviations from the target for each criterion maintenance decision making: the state of the art. Reliab Eng Syst Safe
(i.e. dCK; dCC are the negative and positive 1998;60(2):111–9.
848 M. Bertolini, M. Bevilacqua / Reliability Engineering and System Safety 91 (2006) 839–848

[18] Waeyenbergh G, Pintelon L. Maintenance concept development: a case [29] Lee JW, Kim SH. Using analytic network process and goal programming
study. Int J Prod Econ 2004;89(3):395–405. for interdependent information system project selection. Comput Oper
[19] Kelly A. Maintenance organizations & systems: business-centred Res 2000;27:367–82.
maintenance. Oxford: Butterworth-Heinemann; 1997. [30] Badri MA. Combining the analytic hierarchy process and goal
[20] Almeida AT, Bohoris GA. Decision theory in maintenance decision- programming for global facility location-allocation problem. Int J Prod
making. J Qual Maintenance Eng 1995;1(1):39–45. Econ 1999;62:237–48.
[21] Triantaphyllou E, Kovalerchuk B, Mann L, Knapp GM. Determining the [31] Romero C. Handbook of critical issues in goal programming. Oxford:
most important criteria in maintenance decision making. J Qual Pergamon Press; 1991.
Maintenance Eng 1997;3(1):16–24. [32] Saaty TL. The analytic hierarchy process. New York, NY: McGraw-Hill; 1980.
[22] Bevilacqua M, Braglia M. The analytic hierarchy process applied to [33] Belton V, Stewart TJ. Multiple criteria decision analysis: an integrated
maintenance strategy selection. Reliab Eng Syst Safe 2000;70(1): approach. Dordrecht: Kluwer Academic Publishers; 2002.
71–83. [34] Belton V, Gear AE. The legitimacy of rank reversal - a comment. Omega
[23] Charnes A, Coopper WW. Management models and industrial appli- 1985;13(3):143–4.
cations of linear programming. New York, NY: Wiley; 1961. [35] Vargas LG. An overview of the analytic hierarchy process and its
[24] Kvanli AH. Financial planning using goal programming. Omega 1980;8: applications. Euro J Oper Res 1990;48(1):2–8.
207–18. [36] Saaty TL, Vargas LG. The legitimacy of rank reversal. Omega 1984;
[25] Lin WT. A survey of goal programming applications. Omega 1980;8: 12(5):513–6.
115–7. [37] Saaty TL. An exposition of the AHP in reply to the paper ‘Remarks on the
[26] Taylor III BW, Moore LT, Clayton ER. R&D project selection and man Analytic Hierarchy Process’. Manag Sci 1990;36(3):259–68.
power allocation with integer non-linear goal programming. Manag Sci [38] Saaty TL. That is not the analytic hierarchy process: what the AHP is and
1982;28:1149–58. what it is not. J Multi-Crit Decis Anal 1987;6:324–35.
[27] Schaible S, Karuppan CM. Designing a quality control system in a service [39] Carlos A. Bana e Costa and Jean-Claude Vansnick. A fundamental
organization: a goal programming case study. Eur J Oper Res 1995;81: criticism to Saaty’s use of the eigenvalue procedure to derive priorities,
249–58. 2001, Working Paper LSEOR 01.42, London School of Economics
[28] Badri MA. A combined AHP-GP model for quality control system. Int (available in http://www.lse.ac.uk/collections/operationalResearch/
J Prod Econ 2001;72:27–40. research/workingPapers.htm).

You might also like