Reliability Engineering and System Safety 70 (2000) 7183

www.elsevier.com/locate/ress

The analytic hierarchy process applied to maintenance strategy selection

M. Bevilacqua a, M. Braglia b,*
a
Dipartimento di Ingegneria Industriale, Universita` degli Studi di Parma, Viale delle Scienze, 43100 Parma, Italy
b
Dipartimento di Ingegneria Meccanica, Nucleare e della Produzione, Universita` degli Studi di Pisa, Via Bonanno Pisano 25/B, 56126 Pisa, Italy
Received 12 October 1999; accepted 24 March 2000

Abstract
This paper describes an application of the Analytic Hierarchy Process (AHP) for selecting the best maintenance strategy for an important
Italian oil refinery (an Integrated Gasification and Combined Cycle plant). Five possible alternatives are considered: preventive, predictive,
condition-based, corrective and opportunistic maintenance. The best maintenance policy must be selected for each facility of the plant (about
200 in total). The machines are clustered in three homogeneous groups after a criticality analysis based on internal procedures of the oil
refinery. With AHP technique, several aspects, which characterise each of the above-mentioned maintenance strategies, are arranged in a
hierarchic structure and evaluated using only a series of pairwise judgements. To improve the effectiveness of the methodology AHP is
coupled with a sensitivity analysis. q 2000 Elsevier Science Ltd. All rights reserved.
Keywords: Maintenance; Process plant; Failure mode effect and criticality analysis technique; Analytic hierarchy process

1. Introduction gical choices, managerial and organisational procedures,

Many companies think of maintenance as an inevitable
source of cost. For these companies maintenance operations
have a corrective function and are only executed in emer-
gency conditions. Today, this form of intervention is no
longer acceptable because of certain critical elements such
as product quality, plant safety, and the increase in main-
tenance department costs which can represent from 15 to
70% of total production costs.
The managers have to select the best maintenance policy
for each piece of equipment or system from a set of possible
alternatives. For example, corrective, preventive, opportu-
nistic, condition-based and predictive maintenance policies
are considered in this paper.
It is particularly difficult to choose the best mix of main-
tenance policies when this choice is based on preventive
elements, i.e. during the plant design phase. This is the
situation in the case examined in this paper, that of an Inte-
grated Gasification and Combined Cycle plant which is
being built for an Italian oil company. This plant will
have about 200 facilities (pumps, compressors, air-coolers,
etc.) and the management must decide on the maintenance
approach for the different machines. These decisions will
have significant consequences in the short-medium term for
matters such as resources (i.e. budget) allocation, technolo-
* Corresponding author. Fax: 1 39-050-913-040.
E-mail address: m.braglia@ing.unipi.it (M. Braglia).
0951-8320/00/$ - see front matter q 2000 Elsevier Science Ltd. All rights reserved.
PII: S0951-832 0(00)00047-8
etc. At this level of selection, it is only necessary to define
the best maintenance strategy to adopt for each machine,
bearing in mind budget constraints. It is not necessary to
identify the best solution from among the alternatives that
this approach presents. The maintenance manager only
wants to recognise the most critical machines for a pre-
allocation of the budget maintenance resources, without
entering into the details of the actual final choice. This
final choice would, in any case, be impossible because the
plant is not yet operating and, as a consequence, total
knowledge of the reliability aspects of the plant machines
is not yet available. In other words, the problem is not
whether it is better to control the temperature or the vibra-
tion of a certain facility under analysis, but only to decide if
it is better to adopt a condition-based type of maintenance
approach rather than another type. The second level of deci-
sion making concerns a fine tuned selection of the alterna-
tive maintenance approaches (i.e. definition of the optimal
maintenance frequencies, thresholds for condition-based
intervention, etc.). This level must be postponed until data
from the operating production system becomes available.
Several attributes must be taken into account at this first
level when selecting the type of maintenance. This selection
involves several aspects such as the investment required,
safety and environmental problems, failure costs, reliability
of the policy, Mean Time Between Failure (MTBF) and
Mean Time To Repair (MTTR) of the facility, etc. Several
of these factors are not easy to evaluate because of their
72 M. Bevilacqua, M. Braglia / Reliability Engineering and System Safety 70 (2000) 7183

intangible and complex nature. Besides, the nature of the effectiveness the methodology is coupled with a sensitivity
weights of importance that the maintenance staff must give analysis phase.
to these factors during the selection process is highly subjec-
tive. Finally, bearing in mind that the plant is still in the
construction phase, some tangible aspects such as MTBF
2. The API oil refinery IGCC plant: a brief description
and MTTR can be only estimated from failure data concern-
ing machines working in other plants (in this case oil refi-
The Integrated Gasification and Combined Cycle
neries) under more or less similar operating conditions.
(IGCC) plant [6], currently being assembled at The
Furthermore, they will affect each single facility analysed
Falconara Marittima API oil refinery, will make it
in a particular way and, as a consequence, the final main-
possible to transform the oil refinement residuals into
tenance policy selection.
the synthesis gases which will be used as fuel to
It is therefore clear that the analysis and justification of
produce electricity. The IGCC plan will be placed in
maintenance strategy selection is a critical and complex task
a 47,000 m 2 area inside the oil refinery.
due to the great number of attributes to be considered, many
The electricity produced by the IGCC plant will be sold to
of which are intangible. As an aid to the resolution of this
ENEL (Italian electrical energy firm) while some
problem, some multi-criteria decision making (MCDM)
65,000 ton/h of steam will be used inside the oil refinery
approaches are proposed in the literature. Almeida and
for process requirements. The total cost of the project
Bohoris [1] discuss the application of decision making
amounts to about 750 million dollars.
theory to maintenance with particular attention to multi-
In recent years, economic and legislative changes
attribute utility theory. Triantaphyllou et al. [2] suggest
have led to increased co-operation between petrochem-
the use of Analytical Hierarchy Process (AHP) considering
ical and electrical firms. The adoption of strict environ-
only four maintenance criteria: cost, reparability, reliability
mental standards, both in Europe and in the United
and availability. The Reliability Centered Maintenance
States, is forcing oil refinery firms to reduce the emis-
(RCM) methodology (see, for example, Ref. [3]) is probably
sions of pollutants from the process plants and reduce
the most widely used technique. RCM represents a method
the potential pollution of the refined products. The same
for preserving functional integrity and is designed to mini-
pollution control requirements, mainly a reduction in the
mise maintenance costs by balancing the higher cost of
level of nitrogen and sulphur oxides, together with the
corrective maintenance against the cost of preventive main-
increasing need to control operating and investments
tenance, taking into account the loss of potential life of the
costs, is pushing electrical firms to search for more
unit in question [4].
economic and cleaner production methods.
One of the tools more frequently adopted by the compa-
The combined effect of the above-mentioned factors has
nies to categorise the machines in several groups of risk is
led several oil refineries to adopt IGCC technology for oil
based on the concepts of failure mode effect and criticality
refinement heavy residuals processing. IGCC technology
analysis technique (FMECA). This methodology has been
has proved to be a valid solution to the market requirement
proposed in different possible variants, in terms of relevant
of efficient, clean, low consuming and environmentally
criteria considered and/or risk priority number formulation
orientated production technologies.
[5]. Using this approach, the selection of a maintenance
The API oil refinery uses a thermal conversion process
policy is performed through the analysis of obtained priority
and has a production capacity of about 4,000,000 tons of oil
risk number. An example of this approach has also been
per year (80,000 barrels per day). The production cycle is
followed by our oil company, which has developed its
typical of oil refineries with a similar production capacity:
own methodology internally. This approach makes it
the current distilled yield is higher than 70% and the resi-
possible to obtain a satisfying criticality clustering of
duals are used to produce fuel oil and bitumen. Oil refine-
the 200 facilities into three homogeneous groups. The
ment heavy residuals with a high sulphur content will be
problem is to define the best maintenance strategy for
partly converted into the synthesis gases syngas (which
each group.
will be cleaned in the IGCC gasifiers) and partly used to
To integrate the internal self-made criticality approach,
produce bitumen.
this paper presents a multi-attribute decision method based
The three main objectives of the oil refinery management
on the AHP approach to select the most appropriate main-
are the following:
tenance strategy for each machine group. In this procedure,
several costs and benefits for each alternative maintenance 1. the elimination of heavy residuals used to produce fuel
strategy are arranged in a hierarchic structure and evaluated, oil with high and low sulphur content;
for each facility, through the use of a series of pairwise 2. the ability to process almost every type of heavy oil with
judgements. Finally, considering that the maintenance a high sulphur content;
manager can never be sure about the relative importance 3. the substitution of the present low efficiency thermoelec-
of decision making criteria selected when dealing with trical power plant with a more efficient system, with
this complex maintenance problem, to improve the AHP lower levels of pollutant emissions.
 M. Bevilacqua, M. Braglia / Reliability Engineering and System Safety 70 (2000) 7183
3. Possible alternative maintenance strategies
Five alternative maintenance policies are evaluated in this
case study. Briefly, they are the following.
Corrective maintenance. The main feature of corrective
maintenance is that actions are only performed when a
machine breaks down. There are no interventions until a
failure has occurred.
Preventive maintenance. Preventive maintenance is
based on component reliability characteristics. This
data makes it possible to analyse the behaviour of the
element in question and allows the maintenance engineer
to define a periodic maintenance program for the
machine. The preventive maintenance policy tries to
determine a series of checks, replacements and/or
component revisions with a frequency related to the fail-
ure rate. In other words, preventive (periodic) mainte-
nance is effective in overcoming the problems
associated with the wearing of components. It is evident
that, after a check, it is not always necessary to substitute
the component: maintenance is often sufficient.
Opportunistic maintenance. The possibility of using
opportunistic maintenance is determined by the nearness
or concurrence of control or substitution times for differ-
ent components on the same machine or plant. This type
of maintenance can lead to the whole plant being shut
down at set times to perform all relevant maintenance
interventions at the same time.
Condition-based maintenance. A requisite for the appli-
cation of condition-based maintenance is the availability
of a set of measurements and data acquisition systems to
monitor the machine performance in real time. The
continuous survey of working conditions can easily and
clearly point out an abnormal situation (e.g. the exceed-
ing of a controlled parameter threshold level), allowing
the process administrator to punctually perform the
necessary controls and, if necessary, stop the machine
before a failure can occur.
Predictive maintenance. Unlike the condition-based
maintenance policy, in predictive maintenance the
acquired controlled parameters data are analysed to find
a possible temporal trend. This makes it possible to
predict when the controlled quantity value will reach or
exceed the threshold values. The maintenance staff will
then be able to plan when, depending on the operating
conditions, the component substitution or revision is
really unavoidable.
4. The IGCC plant maintenance program definition
An electrical power plant based on IGCC technology is a
very complex facility, with a lot of different machines and
equipment with very different operating conditions. Decid-
ing on the best maintenance policy is not an easy matter,
73
since the maintenance program must combine technical
requirements with the firms managerial strategy. The
IGCC plant complex configuration requires an optimal
maintenance policy mix, in order to increase the plant avail-
ability and reduce the operating costs.
Maintenance design deals with the definition of the best
strategies for each plant machine or component, depending
on the availability request and global maintenance budget.
Every component, in accordance with its failure rate, cost
and breakdown impact over the whole system, must be
studied in order to assess the best solution; whether it is
better to wait for the failure or to prevent it. In the latter
case the maintenance staff must evaluate whether it is better
to perform periodic checks or use a progressive operating
conditions analysis.
It is clear that a good maintenance program must define
different strategies for different machines. Some of these
will mainly affect the normal operation of the plant, some
will concern relevant safety problems, and others will
involve high maintenance costs. The overlapping of these
effects enables us to assign a different priority to every plant
component or machine, and to concentrate economic and
technical efforts on the areas that can produce the best
results.
One relevant IGCC plant feature is the lack of historical
reliability and maintenance costs data (the plant start-up is
proposed for March 2000). Initially, the definition of the
maintenance plan will be based upon reliability data from
the literature and on the technical features of the machines.
This information will then be updated using the data
acquired during the working life of the plant. The analysis
system has been structured in a rational way so as to keep
the update process as objective as possible. This has been
accomplished through the use of a charting procedure, using
well-understood evaluations of different parameters and a
simple and clear analysis of corrective interventions. The
maintenance plan developed for the machines of the IGCC
plant is based on the well-known FMECA technique [7,8].
The analysis results have provided a criticality index for
every machine, allowing the best maintenance policy to
be selected.
4.1. The maintenance strategy adopted by the oil refinery
company
The internal methodology developed by the company to
solve the maintenance strategy selection problem for the
new IGCC plant is based on a criticality analysis which
may be considered as an extension of the FMECA techni-
que. This analysis takes into account the following six
parameters:
safety;
machine importance for the process;
maintenance costs;
failure frequency;
downtime length;
74 M. Bevilacqua, M. Braglia / Reliability Engineering and System Safety 70 (2000) 7183
Table 1
Weight values assigned to the relevant parameters considered in FMECA
analysis
Parameter Weight
Safety 1.5
Machine importance for the process 2.5
Maintenance costs 2 To restrict the complexity (and the costs) of the analysis
Failure frequency 1
Downtime length 1.5
Operating conditions 1 reduced by grouping together those that are similar and by
operating conditions;
with an additional evaluation for the
machine access difficulty.
Note that, the six parameters presented below derived
from an accurate pre-analysis to select all of the relevant
parameters that can contribute to the machine criticality. As
reported by the maintenance manager, 12 criteria have initi-
ally been considered:
a. Safety. Considering the safety of personnel, equipment,
the buildings and environment in the event of a failure.
b. Machine importance for the process. The importance
of the machine for the correct operation of the plant. For
instance, the presence of an inter-operational buffer to
stock the products can reduce the machine criticality
since the maintenance intervention could be performed
without a plant shutdown.
c. Spare machine availability. Machines that do not have
spares available are the most critical.
d. Spare parts availability. The shortage of spare parts
increases the machine criticality and requires a replenish-
ment order to be issued after a failure has occurred.
e. Maintenance cost. This parameter is based on
manpower and spare parts costs.
f. Access difficulty. The maintenance intervention can be
difficult for machines arranged in a compact manner, placed
in a restrict area because they are dangerous, or situated at a
great height (for example, some agitators electric motors
and air-cooler banks). The machine access difficulty
increases the length of downtime and, moreover, increases
the probability of a failure owing to the fact that inspection
teams cannot easily detect incipient failures.
g. Failure frequency. This parameter is linked to the mean
time between failures (MTBF) of the machine.
h. Downtime length. This parameter is linked to the mean
time to repair (MTTR) of the machine.
i. Machine type. A higher criticality level must be assigned
to the machines which are of more complex construction.
These machines are also characterised by higher mainte-
nance costs (material and manpower) and longer repair
times.
l. Operating conditions. Operating conditions in the
presence of wear cause a higher degree of machine
criticality.
m. Propagation effect. The propagation effect takes into
account the possible consequences of a machine failure
on the adjacent equipment (domino effect).
n. Production loss cost. The higher the machine importance
for the process, the higher the machine criticality due to a
loss of production.
to be performed, the number of evaluation parameters is
removing the less meaningful ones. An increase in the
number of parameters does not imply a higher degree of
analysis accuracy. With a large number of parameters the
analysis becomes much more onerous in terms of data
required and elaboration time. Besides, the quantitative
evaluation of the factors described is complex and subject
to the risk of incorrect estimates. The following clusters
were created.
The spare machine availability mainly affects the unin-
terrupted duration of the production process and can there-
fore be linked to the machine importance for the process
and the production loss cost. In terms of spare parts, the
maintenance cost can include the machine type factor,
while the manpower contribution to the maintenance cost
can be clustered with the downtime length attribute.
System safety, failure frequency, access difficulty
and operating conditions are considered to be stand-
alone factors by the maintenance staff.
For every analysed machine of the new IGCC plant, a
subjective numerical evaluation is given adopting a scale
from 1 to 100. Finally, the factors taken into consideration
are linked together in the following criticality index CI:
CI S 1:5 1 IP 2:5 1 MC 2 1 FF 1
1 DL 1:5 1 OC 1 AD 1
where S safety, IP machine importance for the process,
MC maintenance costs, FF failure frequency, DL
downtime length, OC operating conditions, AD
machine access difficulty.
In the index, the machine access difficulty has been
considered by the management to be an aggravating aspect
as far as the equipment criticality is concerned. It is there-
fore suitable to evaluate the effect of the machine access
difficulty as an a posteriori factor. For this reason with
this approach the machine criticality index has been multi-
plied by the machine access difficulty.
A rational quantification of the seven factors has been
defined and based on a set of tables. In particular, every
relevant factor is divided into several classes that are
assigned a different score (in the range form 1 to 100) to
take into account the different criticality levels. For the sake
of brevity, only the evaluation of the machine importance
for the process attribute is reported as an example in
Appendix A.
 M. Bevilacqua, M. Braglia / Reliability Engineering and System Safety 70 (2000) 7183
Table 2
Maintenance policy selection based on criticality index
Criticality index Maintenance policy
. 395 Predictive
290395 Preventive
, 290 Corrective
The weighted values assigned by the maintenance staff to
the different parameters are shown in Table 1.
The weight assigned to safety is not the highest because in
an IGCC plant danger is intrinsic to the process. The oper-
ating conditions are weighted equal to one in accordance
with the hypothesis of a correct facility selection as a func-
tion of the required service. The breakdown frequency is
weighted equal to one in virtue of the fact that failure
rates are currently estimated values only.
The CI index has been used to classify about 200
machines of the plant (pumps, compressors, air coolers,
etc.) into three different groups corresponding to three
different maintenance strategies, as shown in Table 2.
Note that only corrective, preventive and predictive main-
tenance strategies have been taken into account by the refin-
ery maintenance management.
The main features of the three groups are the following:
Group 1. A failure of group 1 machines can lead to
serious consequences in terms of workers safety, plant
and environmental damages, production losses, etc.
Significant savings can be obtained by reducing the fail-
ure frequency and the downtime length. A careful
maintenance (i.e. predictive) can lead to good levels
of company added-value. In this case, savings in
maintenance investments are not advisable. This
group contains about the 70% of the IGCC machines
examined.
Group 2. The damages derived from a failure can be
serious but, in general, they do not affect the external
environment. A medium cost reduction can be
obtained with an effective but expensive mainte-
nance. Then an appropriate cost/benefit analysis
must be conducted to limit the maintenance invest-
ments (i.e. inspection, diagnostic, etc.) for this type
of facilities (about the 25% of the machines). For this
reason a preventive maintenance is preferable to a
more expensive predictive policy.
Group 3. The failures are not relevant. Spare parts
are not expensive and, as a consequence, low levels
of savings can be obtained through a reduction of
spare stocks and failure frequencies. With a tight
budget the maintenance investments for these types
of facilities should be reduced, also because the
added-value derived from a maintenance plan is
negligible. The cheapest corrective maintenance is,
therefore, the best choice. Group 3 contains 5% of
the machines.
75
4.2. Critical analysis of oil company maintenance MCDM
methodology
Some aspects of the criticality index CI proposed and
prepared by the maintenance staff are open to criticism.
Eq. (1) represents a strange modified version of the
weighted sum model (WSM), which probably represents
the simplest and still the most widely used MCDM method
[2]. But, in this case, there are some weaknesses.
(a) The WSM is based on the additive utility supposi-
tion [2]. However, the WSM should be used only when
the decision making criteria can be expressed in identical
units of measure.
(b) The AD factor should be added and not used as a
multiplying factor.
(c) Dependencies among the seven attributes should be
carefully analysed and discussed.
(d) The weight values reported in Table 1 are not justified
in a satisfying manner. The maintenance staff also have
serious doubts about these values, which would suggest
that they have little confidence in the final results
obtained by the MCDM model. Moreover, no sensibility
analyses have been conducted to test the robustness of the
results. This fact is probably due to (i) a sensitivity analy-
sis is not an easy matter, and (ii) the absence of a software
package supporting this request.
Despite these problems, the classification produced using
the CI index has made it possible to define three homoge-
neous groups of machines. The composition of the clusters
confirms the expectations of the maintenance staff and is
considered to be quite satisfactory. On the other hand, the
doubts of the maintenance staff mainly concern the main-
tenance strategy to adopt for each group of machines. This
factor has been used as the starting point for the develop-
ment of an AHP approach to assign the best maintenance
strategy to each cluster element, taking into account several
possible aspects.
5. The analytic hierarchy process
The AHP [911] is a powerful and flexible multi-criteria
decision making tool for complex problems where both
qualitative and quantitative aspects need to be considered.
The AHP helps the analysts to organise the critical aspects
of a problem into a hierarchical structure similar to a family
tree. By reducing complex decisions to a series of simple
comparisons and rankings, then synthesising the results, the
AHP not only helps the analysts to arrive at the best deci-
sion, but also provides a clear rationale for the choices
made.
Briefly, the step-by-step procedure in using AHP is the
following:
1. define decision criteria in the form of a hierarchy of
objectives. The hierarchy is structured on different levels:
76 M. Bevilacqua, M. Braglia / Reliability Engineering and System Safety 70 (2000) 7183
Table 3
Judgement scores in AHP
Judgement Explanation
Equally Two attribute contribute equally to the upper-
level criteria
Moderately Experience and judgement slightly favour one
attribute over another
Strongly Experience and judgement strongly favour one
attribute over another
Very strongly An attribute is strongly favoured and its
dominance demonstrated in practice
Extremely The evidence favouring one attribute over
another is of the highest possible order of
affirmation
from the top (i.e. the goal) through intermediate levels
(criteria and sub-criteria on which subsequent levels
depend) to the lowest level (i.e. the alternatives);
2. weight the criteria, sub-criteria and alternatives as a func-
tion of their importance for the corresponding element of
the higher level. For this purpose, AHP uses simple pair-
wise comparisons to determine weights and ratings so
that the analyst can concentrate on just two factors at
one time. One of the questions which might arise when
using a pairwise comparison is: how important is the
maintenance policy cost factor with respect to the
maintenance policy applicability attribute, in terms of
the maintenance policy selection (i.e. the problem
goal)? The answer may be equally important, moder-
ately more important, etc. The verbal responses are then
quantified and translated into a score via the use of
discrete 9-point scales (see Table 3);
3. after a judgement matrix has been developed, a priority
vector to weight the elements of the matrix is calculated.
This is the normalised eigenvector of the matrix.
The AHP enables the analyst to evaluate the goodness of
judgements with the inconsistency ratio IR. The judgements
can be considered acceptable if I R # 0:1: In cases of incon-
sistency, the assessment process for the inconsistent matrix
is immediately repeated. An inconsistency ratio of 0.1 or
more may warrant further investigation. For more details on
IR, the reader may refer to Saaty [9].
After its introduction by Saaty [9], AHP has been widely
used in many applications [12]. Designed to reflect the way
people actually think, the AHP was developed more than 20
years ago and it is one of the most used multi-criteria deci-
sion-making techniques. This fact is probably due to some
important aspects of AHP such as [13]: (1) the possibility to
measure the consistency in the decision makers judgement;
(2) it does not, in itself, make decisions but guides the
analyst in his decision making; (3) it is possible to conduct
a sensitivity analysis; (4) AHP can integrate both
quantitative and qualitative information; and (5) it is well
supported by commercial software programs. However, it is
Score necessary to remember some possible problems with AHP
(rank reversal, etc.) which are currently debated in the litera-
1 ture [1416].
2
3
6. The hierarchy scheme
4
5 In designing the AHP hierarchical tree, the aim is to
develop a general framework that satisfies the needs of the
6
7 analysts to solve the selection problem of the best mainte-
nance policy. The AHP starts by breaking down a complex,
8 multi-criteria problem into a hierarchy where each level
9 comprises a few manageable elements which are then
broken down into another set of elements. Considering the
critical aspect of this step for the AHP methodology, the
structure has been created following suggestions from the
maintenance and production staff of the oil refinery (Fig. 1).
The AHP hierarchy developed in this study is a five-level
tree in which the top level represents the main objective of
maintenance selection and the lowest level comprises the
alternative maintenance policies.
The evaluation criteria that influence the primary goal are
included at the second level and are related to four different
aspects: damages produced by a failure, applicability of the
maintenance policy, added-value created by the policy, and
the cost of the policy. These criteria are then broken down
into several sub-criteria as shown in Fig. 1.
The definition of the hierarchy scheme described above
has been developed using a brainstorming process. All the
people concerned with maintenance problems in the oil
refinery where the IGCC plant is being built have been
included in the study. In particular, we have involved the
maintenance engineering personnel (who perform the criti-
cality analyses and develop the maintenance improvement
procedures), the MON (Maintenance On Site) and MOF
(Maintenance Off Site) staff (who manage the maintenance
operations for the transfer and the mixing of the oil refinery
products).
The hierarchy scheme has been based on the fees due
during the oil refinery maintenance management. In this
way the four criteria of the first level of the hierarchy
have been immediately identified. Possible differences of
opinion expressed by maintenance supervisors have been
smoothed using the Delphi technique. In this way we
have tried to maximise the advantages of group dynamics
and reduce the negative effects due to the subjectivity of
opinions.
The formal tool used during the hierarchy structure defi-
nition was the Interpretive Structural Modelling (IMS)
[17,18] approach. IMS is a well-established interactive
learning process for identifying and summarising relation-
ships among specific factors of a multi-criteria decision-
making problem, and provides an efficient means by
which a group of decision-makers can impose order on
 M. Bevilacqua, M. Braglia / Reliability Engineering and System Safety 70 (2000) 7183 77

Fig. 1. The AHP hierarchy scheme adopted.

the complexity of the problem. The steps of this methodol-
ogy are briefly the following [18]:
a. identification of elements which are relevant to the
decision making problem;
b. a contextually relevant subordinate relation is chosen;
c. a structural self-interaction matrix (SSIM) is developed
based on pairwise comparison of elements;
d. SSIM is converted into a reachability matrix and its
transitivity is checked;
e. once the transitivity has been achieved, the conversion
of an object system into a well-defined matrix model is
obtained;
f. the partitioning of the elements and the extraction
of the structural model, termed ISM is then
performed.
The relevant factors defining the damage criteria are iden-
tified as loss of production, damages to facilities, to product,
to environment, to people, and to company image. The
production loss is linked to facility downtime derived
from a failure (time required for detection, repair or repla-
cement and re-starting) and divided in MTBF and MTTR in
a successive hierarchy level. The facility damages are
divided into direct and indirect: the direct damage deals
78 M. Bevilacqua, M. Braglia / Reliability Engineering and System Safety 70 (2000) 7183
Table 4
Pairwise comparison example with respect to damages attribute for machines of Group 1. Note: row element is x (or 1/x) times more (or less) important than
column element
Production loss Plant damage
Production loss 1/7
Plant damage
Product damage
Environmental damage
People damage
Image damage
with the tangible effects of the failure on the machine,
the indirect damage takes into account the possible
influences (i.e. reduction) of the failure on the working
life of the plant as a consequence of a domino effect
on other facilities and instruments. Finally, the environ-
mental damage is divided into internal and external to
the plant.
For the maintenance policy applicability factor, two
sub-criteria are taken into account: the costs required
for the policy implementation and the reliability. The
maintenance costs are divided into hardware (i.e.
sensors), software (i.e. a reliability commercial soft-
ware), and training costs. The reliability criterion
includes the technical aspect of the maintenance strat-
egy adopted in terms of fault detection capability and
facility restoring capability.
The added-value criterion deals with the indirect benefit
of a particular maintenance policy. This category includes
the improvements in terms of product quality, safety, and
internal skills (i.e. an overall better knowledge of the
machines employed).
With respect to the policy cost factor, four sub-
criteria have been considered for the successive hierar-
chy level: MTBF, MTTR, saving in the stock of spare
parts, and assurance aspects (i.e. possible decreases in
insurance premiums that can be obtained by adopting a
particular maintenance policy). Note that the investment
required to implement the policy is not included as sub-
criterion. This is due to the fact that the maintenance
investment for a single machine is negligible with
respect to the other sub-criteria of the policy cost
factor. For this reason, the investment required attri-
bute has been introduced only as sub-criterion of the
policy applicability factor in the hierarchy.
The proposed hierarchy must not be intended as a
general for any process plant application, with some
choices concerning only the particular IGCC plant
here analysed. Besides, it is possible to note how
some possible dependencies among the attributes have
not been treated. These structure simplification choices
derive from the necessity to obtain a good trade-off
Fig. 2. Global priority indices obtained for a Group 1 facility.
Product damage Environmental damage People damage Image damage Local Priority
3 1/5 1/7 1 0.050
9 3 1 9 0.352
1/7 1/9 1 0.030
1/3 7 0.181
9 0.352
0.035
between a suitable level of detail and a manageable
complexity of the hierarchy structure for actual
industrial applications.
7. The AHP analysis
Once the hierarchy structure of the maintenance decision
making problem has been defined, the priorities of the
scheme have been calculated using pairwise comparisons
and the Delphi technique.
The three machine groups previously described have
been analysed through the use of the AHP methodology.
In this situation five different alternative maintenance poli-
cies have been considered. The pairwise judgements used to
perform the AHP analysis were stated by the oil refinery
maintenance management that developed the modified
FMECA analysis shown in Section 4.
Table 4 gives an example of the damages factor for the
machines of Group 1, with the relative judgements of impor-
tance (IR results equal to 0:03 , 0:1 obtained by adopting
the quantitative scale showed in Table 3.
It is clear that the most important aspect, in terms of
damages, is the plant damage attribute. This consid-
eration is due to the fact that the highest cost derived
from an accident is due to production plant damages
(direct and indirect, i.e. domino effect). Environmen-
tal damage is the least important element because of the
presence of modern and sophisticated safety systems
that guarantee limited consequences for the environ-
ment.
Fig. 2 shows the global priority indices for each criterion,
sub-criterion and alternative included in the AHP hierarchy
structure for a particular Group 1 machine (a 23MW power
air compressor).
As can be seen, the overall inconsistency index is equal to
0.046, which is less than the critical value of 0.1. For brev-
ity, the partial IR values are not shown in the figure. Never-
theless, all the IR values result lower than 0.1.
Table 5 shows the final ranking for three machines, each
representative of a different group.
M. Bevilacqua, M. Braglia / Reliability Engineering and System Safety 70 (2000) 7183 79
80 M. Bevilacqua, M. Braglia / Reliability Engineering and System Safety 70 (2000) 7183
As shown, the AHP results generally confirm the FMECA
indications. Once again, for the Group 1 facility the best
maintenance policy proves to be the predictive one. For
these machines the most expensive and sophisticated stra-
tegies (i.e. predictive and condition-based) are highly
preferable with respect to the others because of the critical
aspect of their failures.
The Group 2 machine shows a slight preference for
opportunistic maintenance even if preventive and predictive
policies should not be neglected.
The analysis performed for the machine belonging to
Group 3 shows that corrective and preventive maintenance
are equally appropriate. So, the decision to use corrective
maintenance for this group of machines should be carefully
considered. The limited failure consequence in terms of
production loss is not sufficient to clearly and immediately
state that no maintenance action should be performed before
the failure.
The three machines selected for the AHP analysis can be
considered as representative of the corresponding groups. In
other words, they represent more or less the average char-
acteristics of the facilities which belong to the same cluster.
For this reason, the results obtained for this representative
equipment have been extended to all facilities of the groups.
This position has been considered sufficient and satisfactory
by the maintenance staff. In any case, a more detailed analy-
sis requiring a new set of pairwise judgements for each
facility of each group can be always performed. This is
probably the best solution for the extreme equipment
characterised by the CI values near to the limit of the
belonging cluster.
8. Sensitivity analysis
Although the ranking solution showed a possible scenario
where, for example, damages aspects are clearly the most
important criteria for Group 1 machines (see Fig. 2), the
AHP solution can change in accordance with shifts in
analyst logic. To explore the response of model solutions
(i.e. the solution robustness) to potential shifts in the priority
of maintenance strategies, a series of sensitivity analyses of
criteria weights can be performed by changing the priority
(relative importance) of weights. Each criterion can be char-
acterised by an important degree of sensitivity, i.e. the rank-
ing of all strategies changes dramatically over the entire
weight range [19]. The problem is to check whether a few
changes in the judgement evaluations can lead to significant
modifications in the priority final ranking. For this reason,
sensitivity analysis is used to investigate the sensitivity of
the alternatives to changes in the priorities of the criteria at
the level immediately below the goal. The analysis proposed
emphasises the priorities of the four first-level criteria in
the AHP model reported in Fig. 1 and shows how changing
the priority of one criterion affects the priorities of the
others. It is clear that as the priority of one of the criteria
increases, the priorities of the remaining criteria must
decrease proportionately, and the global priorities of the
alternatives must be recalculated. All the results reported
in Tables 68 were obtained using the Expertchoice soft-
ware, a multi-attribute decision making tool which was used
to support all the AHP applications reported in this paper.
For example, from Table 6, one can clearly see the
robustness of the predictive maintenance strategy for the
machines that belong to Group 1. This policy remains the
best solution, increasing or decreasing the priorities of each
criterion. It is only when one increases the priority of the
applicability factor to an improbable 96% that we obtain a
change in the final ranking with the condition-based main-
tenance strategy.
For the Group 2 machines (Table 7), the final solution is
less stable. Opportunistic and predictive maintenance stra-
tegies show some alternations in priority positions. In parti-
cular, the maintenance staff must take into consideration
applicability and added-value priority evaluations.
There are no situations where catastrophic changes in the
final ranking are observed.
An analysis of the Group 3 machines (Table 8), highlights
more reversals in the final ranking. Corrective, opportunistic
and predictive maintenance strategies can all be considered
to be the best solution, depending on which weighting
criteria are used. It can, therefore, be seen that the sensitivity
analysis shows how, in the case study under examination,
the maintenance staff should concentrate their attention on
the pairwise evaluation of the four criteria. It is clear that a
few changes in the judgement evaluations can lead to modi-
fications in the final priority ranking. But it must also be
remembered that the importance of corrective maintenance
never goes away. In other words, we have three different
strategies, all of which can provide the best solution, and the
differences among them are never of great importance.
Summarising the considerations discussed above, one can
affirm that the AHP selections presented in Section 7, which
concern the best possible maintenance strategies for the
IGCC machines, can be accepted with a good degree of
confidence by the company maintenance staff.
We conclude this section with two final considerations.
First, the sensitivity analysis proposed here is only relevant
to the priorities of the four first-level criteria. Second,
because we have changed each attribute weight one at a
time, only the main effects have been considered. In
other words, interaction effects of the changes made to
two or more weights have been ignored. These simplifica-
tions have been adopted for the following reasons:
1. the final solution is mainly sensible to changes in the
priorities at the highest level of the hierarchy;
2. the introduction of the interaction effects makes the
sensitivity analysis too complex for actual applications.
Nevertheless, one should note that the main effects are
generally the most important aspects in a sensitivity
analysis.
 M. Bevilacqua, M. Braglia / Reliability Engineering and System Safety 70 (2000) 7183 81

In other words, the simple approach described here seems

Corrective maintenance
to be a good compromise between efficiency and efficacy.

9. Conclusions

0.043
0.174
0.227
The definition of the most appropriate maintenance poli-
cies for a large system such as an IGCC plant requires the
development of the appropriate decision support systems.
Preventive maintenance

The maintenance plan selection for each component is very

complex due to the difficulties concerning data collection,
the number of factors to be taken into account, their subjec-
tivity, the large number of the plant machines (around two
hundred in the case under examination) to be considered,
0.120
0.212
0.224

and the fact that the plant is still being built.

When integrated with a pre-analysis of the facility criti-
cality, the AHP technique has proved to be a valid support
for the selection of the maintenance strategy. The hierarch-
Opportunistic maintenance

ical structure of the proposed AHP combines many features

which are important for the selection of the maintenance
policy: economic factors, applicability and costs, safety,
etc. Also, accepting the fact that there are some differences,
the AHP results are not completely different from those
directly proposed by oil refinery personnel. This is not a
0.136
0.219
0.219

surprise because the same experts performed the two

analyses. But the AHP approach is characterised by some
important properties which are considered highly by the
Condition based maintenance

maintenance staff:
a. AHP technique makes it possible to approach the deci-
sion making problem in a more complete and thorough
way, taking several factors into account. This capacity is
more difficult to obtain when using conventional meth-
odologies such as FMECA. It must also be considered
AHP final ranking for three machines, each one representative of a different group

0.333
0.195
0.165

that the AHP is able to manage a large number of possible

alternatives in an efficient way;
b. AHP can integrate both qualitative and quantitative
information. With AHP a direct quantitative judgement
Predictive maintenance

of the relevant maintenance factors is not necessarily

required by the maintenance manager. The pairwise
comparisons are preferred by the manager when several
intangible criteria have to be treated, as is the case with
maintenance selection. In addition, AHP is the only
0.368
0.199
0.165

known MCDM model that can measure the consistency

of the decision maker;
c. AHP procedure is readily available in decision-making
software packages from several commercial software
Machine description

Centrifugal pump
Centrifugal pump

sellers. In particular, these software tools make it possible

Air compressor

to conduct a sensitivity analysis which improves the

effectiveness of the procedure and the stability of deci-
sions reached through AHP.
The results and the satisfaction of maintenance manage-
ment derived by using the proposed methodology confirms
how AHP can enhance and improve the understanding of the
Group 1
Group 2
Group 3
Table 5

dynamics of a similar complex problem and represents an

effective approach to arrive at decisions.
82 M. Bevilacqua, M. Braglia / Reliability Engineering and System Safety 70 (2000) 7183

Table 6
Observations derived from sensitivity analysis of the criteria priority values with machines of Group 1

Criteria Decreasing Relative priority value Increasing

in final solution (%)

Damages Predictive maintenance is always the best 56.6 Predictive maintenance is always the best
strategy strategy
Applicability Predictive maintenance is always the best 36.8 Predictive maintenance strategy is reached
strategy and overcame by condition-based
maintenance but only when applicability
priority is equal to 96%.
Added-value Predictive maintenance is always the best 36.8 Predictive maintenance is always the best
strategy strategy
Cost Predictive maintenance is always the best 26.7 Predictive maintenance is always the best
strategy strategy

Table 7
Observations derived from sensitivity analysis of the criteria priority values with machines of Group 2

Criteria Decreasing Relative priority value Increasing

in final solution (%)

Damages Opportunistic maintenance is always the 26.2 Opportunistic maintenance strategy is

best strategy reached and overcame by predictive
maintenance when damages priority is
equal to 34.2%.
Applicability Opportunistic maintenance strategy is 56.5 Opportunistic maintenance is always the
reached and overcame by predictive best strategy
maintenance just when applicability
priority is reduced to 50.5%.
Added-value Opportunistic maintenance is always the 5.5 Opportunistic maintenance strategy is
best strategy reached and overcame by predictive
maintenance when added-value priority is
equal to 11.5%.
Cost Opportunistic maintenance is always the 11.8 Opportunistic maintenance strategy is just
best strategy reached and overcame by predictive
maintenance when cost priority is equal
to 19.8%.

Table 8
Observations derived from sensitivity analysis of the criteria priority values with machines of Group 3

Criteria Decreasing Relative priority value Increasing

in final solution (%)

Damages Corrective maintenance is always the best
strategy
Applicability Corrective maintenance is reached and
overcame by opportunistic maintenance just
when applicability priority is reduced to
61.5%. Predictive is the new best strategy
when priority is reduced to 47.5%
Added-value Corrective maintenance is always the best
strategy
Cost Corrective maintenance is always the best
strategy
15.1 Corrective maintenance is reached and
overcame by opportunistic maintenance just
when damages priority is became to
17.1%. Predictive is the new best strategy
when priority is equal to 53.1%
63.5 Corrective maintenance is always the best
strategy
6.2 Corrective maintenance is reached and
overcame by opportunistic maintenance
when damages priority is became to
10.2%. Predictive is the new best strategy
when priority is equal to 20.2%
11.8 Corrective maintenance is reached and
overcame by opportunistic maintenance
when damages priority is became to
19.1%. Predictive is the new best strategy
when priority is equal to 33.1%
 M. Bevilacqua, M. Braglia / Reliability Engineering and System Safety 70 (2000) 7183 83

Table A1 making. Journal of Quality in Maintenance Engineering

Scores for the evaluation of machine importance for the process factor 1995;1(1):3945.
[2] Triantaphyllou E, Kovalerchuk B, Mann L, Knapp GM. Determining
Impact on the process With spare machine Without spare machine the most important criteria in maintenance decision making. Journal
of Quality in Maintenance Engineering 1997;3(1):1624.
Low 10 50 [3] Rausand M. Reliability centered maintenance. Reliability Engineer-
Medium 30 70
ing and System Safety 1998;60:12132.
High 70 90 [4] Crocker J, Kumar UD. Age-related maintenance versus reliability
Very high 80 100
centered maintenance: a case study on aero-engines. Reliability Engi-
neering and System Safety 2000;67:1138.
[5] Gilchrist W. Modelling failure modes and effects analysis. Interna-
Acknowledgements tional Journal of Quality and Reliability Management 1993;10(5):16
23.
We would like to thank the referees for their constructive [6] Del Bravo R, Starace F, Chellini IM, Chiantore PV. Impianto IGCC
comments that enabled us to improve the quality of this da 280 MW di Api Energia in Italia. Lenergia elettrica
1997;74(2):1724.
paper.
[7] Countinho JS. Failure-effect analysis. Transactions NY Academic
Science 1964;26(II):56484.
[8] Holmberg K, Folkeson A. Operational reliability and systematic
Appendix A
maintenance. London: Elsevier, 1991.
[9] Saaty TL. The analytic hierarchy process. New York: McGraw-Hill,
An example of evaluation of a factor of Eq. (1) is reported 1980.
below. The machine importance for the continuity of the [10] Saaty TL. Decision making for leaders. Belmont, CA: Lifetime
IGCC plant operations is evaluated through the analysis of Learning Publications, 1982.
the Process Flow Diagrams (PFDs) and Piping and Instru- [11] Saaty TL. How to make a decision: the analytic hierarchy process.
European Journal of Operational Research 1990;48:926.
mentation Diagrams (P&IDs) of the plant, taking into
[12] Vargas LG. An overview of the analytic hierarchy process and its
account the availability of spare machines. applications. European Journal of Operational Research 1990;48:28.
The scores are assigned in accordance with Table A1, [13] Madu CN, Kuci C-H. Optimum information technology for socio-
which is drawn up with the oil refinery maintenance staff economic development. Information Management and Computer
and process engineers. Security 1994;2(1):411.
Low impact equipment are those items characterised by [14] Belton V, Gear T. On a short-coming of Saatys method of analytic
hierarchies. Omega 1983;11(3):22830.
minor functions or which operate in circuits where it is [15] Dyer JS. Remarks on the analytic hierarchy process. Management
possible to provide the products with intermediate storage. Science 1990;36(3):25968.
In the latter case, the buffer can provide the plant with the [16] Saaty TL. An exposition of the AHP in reply to the paper: remarks on
necessary operating autonomy so that maintenance can be the analytic hierarchy process. Management Science
performed without a plant shut-down. Machines with high 1990;36(3):25968.
[17] Sage AP. Interpretive structural modelling: methodology for lar. New
or very high impact on the process have also been assigned a
York: McGraw-Hill, 1977.
high score for the availability of a spare part to take into [18] Mandal A, Deshmukh SG. Vendor selection using interpretive struc-
account the possible failure of the spare machine to start. tural modelling (IMS). International Journal of Operations and
Production Management 1993;14(6):5259.
[19] Min H, Melachrinoudis E. The relocation of a hybrid manufacturing/
References distribution facility from supply chain perspective: a case study. Inter-
national Journal of Management Science 1999;27:7585.
[1] Almeida AT, Bohoris GA. Decision theory in maintenance decision