Decision Support Systems 38 (2005) 539 – 555

www.elsevier.com/locate/dsw

Selection of diagnostic techniques and instrumentation in a

predictive maintenance program. A case study
M.C. Carnero *
University of Castilla-La Mancha, Technical School of Industrial Engineering, Avda. Camilo Jose Cela s/n, 13071 Ciudad Real, Spain

Received 7 October 2002; received in revised form 22 September 2003; accepted 22 September 2003
Available online 4 November 2003

Abstract

Predictive maintenance programs (PMPs) can provide significant advantages in relation to quality, safety, availability and
cost reduction in industrial plants. Nevertheless, during implementation, different decision making processes are involved, such
as the selection of the most suitable diagnostic techniques. A wrong decision can lead to the failure of the setting up of the
predictive maintenance program and its elimination, with the consequent economic losses, as the setting up of these programs is
a strategic decision. In this article, a model is proposed that carries out the decision making in relation to the selection of the
diagnostic techniques and instrumentation in the predictive maintenance programs. The model uses a combination of tools
belonging to operational research such as: analytic hierarchy process (AHP) and factor analysis (FA). The model has been tested
in screw compressors when lubricant and vibration analyses are integrated.
D 2003 Elsevier B.V. All rights reserved.

Keywords: Predictive maintenance; Decision making; Analytic hierarchy process; Factor analysis

1. Introduction els. This article aims to contribute towards resolving

this problem.
The continuous production process requires a high Although there is a limited number of decision
degree of availability and the elimination of unex- support systems related to predictive maintenance, the
pected breakdown that could cause a prolonged stop- following models should be taken into consideration.
page in production [9]. Predictive maintenance can In Ref. [15], a proportional-hazards model with Wei-
contribute to improving plant availability, safety, bull baseline hazard function and time-dependent
quality, reduction of maintenance costs, etc. This stochastic covariates representing monitored condi-
has led to an increase in the number of predictive tions is suggested and a software is developed to
maintenance programs (PMPs) applied, but, during assist engineers to optimize decisions. In Ref. [30],
the setting up of a PMP, there is a number of decisions Markov models are described for establishing opti-
involved that lack decision support systems or mod- mum inspection intervals for phased deterioration of
monitored complex components in a system with
severe down time costs. In Ref. [16], statistical
* Tel.: +34-926-295-300; fax: +34-926-295-361. analysis of vibration data is undertaken using a
E-mail address: Carmen.Carnero@uclm.es (M.C. Carnero). software package to establish the key vibration signals

0167-9236/$ - see front matter D 2003 Elsevier B.V. All rights reserved.
doi:10.1016/j.dss.2003.09.003
540 M.C. Carnero / Decision Support Systems 38 (2005) 539–555

that are necessary for risk estimation. Ref. [19] The advantages of the introduction of predictive
presents a real-time neural network-based condition maintenance programs (PMPs) are:
monitoring system for rotating mechanical equipment.
In Ref. [29], condition predictors of significant items
Exclusive control of the machines that show the
of the system are monitoring taking into account the beginning of a malfunction.
availability and cost-effectiveness of the monitoring
An increase in the availability of the industrial
techniques. plants [40].
In this article, a model is presented for the
The capacity to carry out quality checks of
selection of diagnostic techniques and instrumenta- both internal and subcontracted maintenance
tion in a predictive maintenance program. To con- interventions.
struct the model, factor analysis and analytic
An increase in the security of the factory [9].
hierarchy process are combined. The model is applied
It facilitates certification and ensures the verifica-
to screw compressors which are monitored by means tion of the requisites of the standard ISO 9000.
of PMPs based on lubricant and vibration analyses
Provides the best programming of maintenance
and when the aforementioned techniques are applied actions.
simultaneously.
Enables the effective programming of supplies and
The layout of the paper is as follows. Section 2 is staff.
an introduction to predictive maintenance techniques,
Production quality is optimized by operating
lubricant and vibration analyses and the integration of machinery without interruption due to failures [21].
both techniques are presented. Section 3 describes the
Support in the design phase of equipment,
characteristics of the mathematical tools used in the particularly by means of the application of modal
construction of the decision support model proposed: analysis [3].
factor analysis and analytic hierarchy process. Section
Reduction of direct maintenance costs by checking
4 presents the model for the selection of diagnostic only the equipment that is developing a fault [38].
techniques and instrumentation in predictive mainte-
By keeping to delivery dates, and by satisfying the
nance. Section 5 describes the application of the customers’ demand for quality, the image of the
model to a screw compressor. Section 6 presents the company is improved.
results obtained from applying the model to a PMP
Costs are brought down in relation to spare parts
integrating lubricant and vibration analyses. Section 7 and labour [18].
presents the conclusions.
By maintaining the industrial equipment opera-
tional whilst applying the predictive tools, the
measuring process does not directly affect the
2. Predictive maintenance techniques: lubricant availability of the equipment.
and vibration analyses
Decrease in the costs related to insurance policies
as security within the factory increases.
Predictive maintenance is a maintenance policy in
Historical information on each piece of equipment
which selected physical parameters associated with an is completed, which helps to determine reliability
operating machine are sensed, measured and recorded parameters and to optimize maintenance planning
intermittently or continuously for the purpose of [2]. This information on the machines and equip-
reducing, analyzing, comparing and displaying the ment is available to the management for decision
data and information so obtained for support decisions making.
related to the operation and maintenance of the
Reduction of energy consumption.
machine [5]. There are numerous predictive techni-
ques, as can be checked in Ref. [12]: lubricant Industrial plants generally possess PMPs based on
analysis, vibration analysis, thermography, penetrat- vibration analysis [4], whereas medium-sized compa-
ing liquids, radiography, ultrasound, control of corro- nies are starting to incorporate them in their Mainte-
sion, etc.; each technique is applied to a type of nance Departments. Their suitability for application to
specific industrial equipment. rotary and reciprocating machines [36], which can be
 M.C. Carnero / Decision Support Systems 38 (2005) 539–555 541

considered to be the most widely used in general, as machinery which requires investigation of the state of
well as their high capacity of diagnosis, make them wear of the equipment, level of oil contamination, and
the most versatile predictive technique. In order to oil condition and includes a recommendation outlin-
carry out the setting up of a PMP based on vibration ing any corrective or preventive maintenance actions
analysis, it is vital to understand their technical that are necessary [39]. In the PMPs based on lubri-
peculiarities, regarding instrumentation, procedures cant analysis dispersion in the information available
and uses that make the production of diagnoses about each test has been detected in managerial and
possible. The success or failure of the setting up laboratory practice and there are no specifications on
process will depend on the program planner’s knowl- the collections of tests that provide the most effective
edge of these subjects. information [26]. It has also been appreciated that
According to Ref. [37], the presence of a fault in PMPs are supported by tests that provide redundant
industrial equipment, whilst still in its incipient phase, information. There is also a shortage of information
will be accompanied by a detectable increase or mod- about the latest generation technology that enables
ification of vibratory signals. There is a wide range of this type of analysis and imprecisions in the industrial
diagnostic techniques that can be applied in vibration plants that try to implement a program also exist [8].
analysis to identify anomalies in machinery. It is All the diagnostic techniques used in a PMP
necessary to investigate which is the most appropriate based on lubricant analysis can have top limits,
technique or techniques for diagnosis in a specific bottom limits or both. The evolution of each param-
machine; the operating conditions that exist, the degree eter can be represented depending on the hours of
of criticality of the machine, the means, as well as the operability of the lubricant. The value of the curve in
personnel available for the control of the analysis, are the analysis of trend not only shows the evolution of
determining factors of the analysis to be performed. the condition of the lubricant and the machine, but
There are two types of PMPs based on vibration also the speed at which the abovementioned trans-
analysis: with portable instrumentation and on-line formation takes place. The intersection of the line
acquisition system. With the portable system, data are that establishes the trend with the value of the limit
acquired at periodic intervals of time and are later that is first reached, whether bottom or top, tells us
downloaded onto a computer [8]. The discontinuous the time that must pass before a state of danger is
character hinders the obtaining of information regard- reached; this characteristic represents the remaining
ing the starting up and stopping of machinery and the life time of the equipment [13].
instants in which the process parameters change. The The diagnostic techniques based on vibration and
costs of this type of PMP are lower than an on-line lubricant analysis that are applied at present appear in
system because the installation of instrumentation is Table 1. There are diagnostic techniques that provide
not needed and the number of sensors can be quantitative data (to which factor analysis will be
reduced. applied) and others that give qualitative information.
In on-line systems, the sensors are fixed in the The integration of lubricant and vibration analyses
measurement position, information being obtained on- can provide significant profits, which so far have not
line of the level of vibrations, which includes infor- been sufficiently analyzed. For this reason, we will
mation of transitory states such as starting up and now go on to detail the most relevant characteristics
stopping [5]. The costs of implementation are very that a predictive maintenance program that integrates
much higher than in the case of portable systems, and the analysis of vibrations and lubricants must possess.
consequently, it is applied to critical machinery or that A PMP integrating vibration and lubricant analysis
found in dangerous environments. involves the acquisition of information of both tech-
The distinction between portable systems and on- niques, in order to correlate all the predictive infor-
line systems will be illustrated in the model proposed mation to obtain an early diagnosis of the root causes
later; different results being obtained in each case. of the failures and the prediction of their consequen-
Lubricant analysis consists of analysis of the state ces on the machinery.
of different physical and chemical parameters of oil in Besides the benefits previously mentioned ob-
order to verify the condition of the lubricant and the tained through the application of a PMP, the integra-
542 M.C. Carnero / Decision Support Systems 38 (2005) 539–555

Table 1
Each of the predictive techniques detects deterio-

Diagnostic techniques in lubricant and vibration analyses [8] ration in different phases of its evolution [25].
Vibration analysis Lubricant analysis Lubricant analysis is capable of detecting the
Spectral analysis Viscosity anomaly in the early phases of its development,
Index of viscosity whereas vibration analysis will only be able to
Analysis of harmonic Water content
evaluate it when the breakdown has already
and orders Total acid number
Trend of the global Total basic number occurred.
value of vibration
Detection of the root causes of the failures [24,35].
Cepstrum Insoluble in pentane
and benzene Consequently, there is an intensification and im-
Analysis of Freezing point
provement of the level of information regarding
temporary signal Igniting point
Spike energy Combustion point incidents that can be transmitted to the personnel in
Demulsibility charge of corrective activities.
Bode plot Tendency to foaming Nevertheless, the simultaneous application of both
Tendency to formation of predictive techniques does not necessarily mean their
coal and ash content
integration. The computer applications that have been
Polar plot
Corrosion to copper sheet reviewed have designed the databases of each of the
Waterfalls Resistance to oxidation predictive technologies on an independent basis and
Colour without establishing any link of union between them
Orbital analysis Stain of oil [23]. Furthermore, the instrumentation used in each of
Point of aniline
the predictive techniques usually comes from different
Statistical analysis Interfacial tension
Dielectric stiffness manufacturers, which also impedes the integration of
Hilbert transform Spectroscopy of atomic emission the information [22].
Envelope Infrared spectroscopy through The capacity of integration of particle counting and
Fourier transform (FTIR) ferrography techniques with spectral analysis is dem-
Modal analysis Ferrografy
onstrated in the processes of wear and pollution that
Particles count
lead to a mechanical breakdown [10]. The condition
of the equipment can be determined by means of
spectral analysis, and ferrographical analysis will
tion of predictive techniques provides the following determine with accuracy the element or component
additional advantages: that is developing the anomaly; particle counting can
also determine the gradient of change of the condition

An increase in the number of pieces of equipment and determine if the initial cause of the failure is due
covered by a PMP. Vibration analysis is generally to an external factor. The periodic use of spectra in
applied in industrial plants, due to the prevalence of waterfall is suggested as a diagnostic technology, as it
rotary and reciprocating machinery, whereas lubri- is a technique that identifies the condition of the
cant analysis has been adopted by machine tool, equipment and the failure that has developed, aspects
maritime and terrestrial fleets and electrical sub- that cannot be investigated by means of other diag-
stations submitted to significant operational loads, nostic techniques; moreover, as it is carried out by
etc. [35]. means of comparison, it restricts the time of identifi-

An increase in the set of anomalies that can be cation of the failure.
controlled. Certain types of damage can only be The application of the diagnostic techniques on an
investigated by means of one of the predictive independent basis is recommended until as much
techniques [17]. knowledge as possible has been obtained using each

Guarantee of the diagnoses provided as the technique individually, in order to later proceed to
information from both techniques is contrasted. their integration. Once this level has been reached, it
All the techniques present deficiencies [5,6,20], and would be convenient to have software available that
it is therefore advisable to confirm the diagnosis. incorporates utilities for this integration.
 M.C. Carnero / Decision Support Systems 38 (2005) 539–555 543
pffiffiffiffi
3. Introduction to the tools factor analysis and If V L is called A (factor loading matrix), then
analytic hierarchy process
R ¼ AAV ð5Þ
3.1. Factor analysis this equation is called the fundamental equation for
FA.
Factor analysis (FA) is a statistical procedure used The matrix A contains correlations between factors
to determine the basic essential variables underlying a and variables. Usually a factor is more interpretable
large number of interrelated variables; a method of when a few variables load highly on it and the rest do
processing data comprising too many variables to not. The factorial matrix indicates the relation be-
allow direct analysis [27]. tween the factors and the variables. Nevertheless, it is
In FA, the researcher is usually interested in dis- often difficult to interpret the factors from the factorial
covering which variables in a data set form a coherent matrix. To facilitate the interpretation the matrix is
subgroup that are relatively independent of one anoth- rotated.
er. The specific goal of analysis may be to outline The rotation consists of turning the axes of coor-
patterns of intercorrelatives among variables, to reduce dinates, which represent the factors, until they are as
a large number of variables to a smaller number of close as possible to the variables in which they are
clusters while retaining maximum spread among ex- saturated. The saturation of factors transforms the
perimental units, (to provide an operational definition initial factorial matrix into another matrix called a
(a regression equation) for an unobserved, hypotheti- factorial rotated matrix, of easier interpretation. The
cal construct by using observed variables, or to test a factorial rotated matrix is a linear combination of the
theory about the nature of underlying variables. first one and explains the same quantity of initial var-
Steps in a FA includes: iance [11]. Rotating is normally used after extracting to
maximise high correlations and minimise low ones.

Selecting and measuring a group of variables. Several methods of rotation exist. The most advis-

Preparing the correlation matrix. able is orthogonal rotation, and of these the most used

Determining the number of components or factors type is the varimax. The varimax technique accom-
to be considered. plishes this aim by means of a transformation matrix

Extracting a set of components or factors from the
[1].
correlation matrix. 0 1

Rotating the components or factors to increase cosc sinc
interpretability.
¼@ A

Interpreting the results. sinc cosc

being w the rotation angle. Then,

The correlation matrix R can be diagonalized
accomplishing and premultiplying it by the matrix V Aunrorated
¼ Arotated ð6Þ
and its transpose VV[1]. R̄ ¼ Aunrotated Aunrotated
V ð7Þ
L ¼ VVRV ð1Þ Rres ¼ R  R̄ ð8Þ
The matrix of eigenvectors V premultiplied by its the elements of this matrix must be small.
transpose produces the identity matrix: Regression coefficients for producing factor scores
from variable scores are a product of the inverse of the
VVV ¼ I ð2Þ correlation matrix and the factor loading matrix
so, reorganising Eq. (1):
B ¼ R1 A ð9Þ
R ¼ VLVV ð3Þ Factor scores are a product of standardised scores
the correlation matrix can be decomposed, on variables and regression coefficients.
pffiffiffiffi pffiffiffiffi
R ¼ ðV LÞð LVVÞ ð4Þ F ¼ ZB ð10Þ
544 M.C. Carnero / Decision Support Systems 38 (2005) 539–555

Standardised scores on variables may be predicted of using an analytic hierarchy process (AHP) is to
as a product of scores on factors weighted by factor identify the preferred alternative and also determine a
loading. ranking of the alternatives when all the decision
criteria are considered simultaneously [33]. The use
Z ¼ FAV ð11Þ of AHP instead of another multicriteria technique is
Correlations among factors may be obtained by due to the following reasons:
producing a matrix of cross products of standardised
Quantitative and qualitative criteria can be in-
factor scores and dividing the results by the number of
cluded in the decision making.
cases minus one.
A large quantity of criteria can be considered.
 
1
A flexible hierarchy can be constructed according
/¼ FFV ð12Þ
N1 to the problem.
The structure matrix C is a product of the pattern With AHP, a complete classification of alternatives
matrix of correlating among factors. can be obtained. Therefore, a hierarchy must be
C ¼ A/ ð13Þ constructed as shown in Fig. 1. In this hierarchy, the
relationship between the goal, criteria, subcriteria and
In the phase of interpretation, the following steps alternatives is established.
are considered to be fundamental: There are three main steps involved in using AHP:

The study of the composition of the significant
The relevant criteria and alternatives must be
factorial saturations of each factor. determined.

The naming of the factors. The name must coincide
Numerical measures must be attached according to
with the structure of the saturations. the relative importance (weights) of the criteria and
the relative performance of the alternatives to these
3.2. Analytic hierarchy process criteria.

The numerical values must be processed in order to
Decision analysis is used when a decision maker determine a ranking of each alternative.
wishes to evaluate the performance of a number of
alternative solutions for a given problem. These alter- In a decision making problem, M alternatives Ai
natives can be evaluated in terms of a number of (i = 1, 2, 3,. . .,M) and N criteria Cj ( j = 1,2,3,. . .,N) are
decision criteria. Often an alternative may be superior considered.
in terms of one or some of the decision criteria, but In order to determine the relative importance of the
inferior in terms of some other criteria. The objective alternatives with regard to each of the criteria or

Fig. 1. AHP hierarchy.

 M.C. Carnero / Decision Support Systems 38 (2005) 539–555 545

between two criteria, linguistic terms are used that When exact measurements of the criteria in a scale
include the judgments of the decision maker. The are available for carrying out the comparisons, that is
linguistic terms are generally associated to numerical to say, w1,w2,. . .,wn, a perfectly consistent matrix is
values constituting a scale [34]. obtained that verifies [28]:
The scale proposed by Saaty is shown in Table 2.
wi
The quantified judgment on pair of criteria Ci and ¼ aij i; j ¼ 1; 2; . . . ; n; ð16Þ
Cj are represented by an N  N matrix A: wj
2 3 From the previous expression, it can be deduced
a11 a12 : : : a1n
6 7 that:
6 7
6 a21 a22 . . . a2n 7 wj
6 7
A¼6 7 ð14Þ aij ¼ 1 i; j ¼ 1; 2; . . . ; n; ð17Þ
6 7 wi
6: : : : : : : : : : : : 7
6 7
4 5 and then:
an1 an2 . . . ann
X
n
wj
where the aij is the relative importance of Ci to Cj. aij ¼ n i ¼ 1; 2; . . . ; n; ð18Þ
j¼1
wi
The quantified judgment between alternatives with
respect to criteria Ci is represented by an M  M or:
matrix.
The following rules must be verified: Xn
aij wj ¼ nwi i ¼ 1; 2; . . . ; n; ð19Þ
j¼1
If aij = a then aji = 1/a, a = 0.
If Ci is judged to be of equal relative importance as and is expressed in its matricial form as [28]:
Cj, then aij = aji = 1, and aii = 1 for all i.
Aw ¼ nw; ð20Þ
If all the comparisons are perfectly consistent, then
the relation: where w is an eigenvector of A with eigenvalue n.
That is to say, since the comparisons matrix pos-
aik ¼ aij ajk bi; j; k: ð15Þ
sesses a range 1, all the eigenvalues are zero except
should always be true for any combination of com- one with value n. The sum of the eigenvalues of a
parisons taken from the judgment matrix. positive matrix is equal to the trace of the matrix, and

Table 2
Scale of relative importances [28]
Intensity of Verbal scale Explanation
importance
1 Equal importance Two activities contribute equally to the objective
3 Weak importance of one over another Experience and judgment slightly favour one activity
over another
5 Essential or strong importance Experience and judgment strongly favour one activity
over another
7 Demonstrated importance An activity is strongly favoured and its dominance
demonstrated in practice
9 Absolute importance The evidence favouring one activity over another is
of the highest possible order of affirmation
2, 4, 6, 8 Intermediate values between the two adjacent judgments When compromise is needed
Reciprocals of If activity i has one of the above (nonzero) numbers –
above numbers assigned to it when compared with activity j, then j has
the reciprocal value when compared with i
546 M.C. Carnero / Decision Support Systems 38 (2005) 539–555

the eigenvalue different to zero is named maximum In the AHP, the pairwise comparisons in a judg-
eigenvalue (kmax). ment matrix are considered to be adequately consis-
If the matrix A is not consistent and k1,. . .,kn is the tent if the corresponding CR is less than 10%. If the
set of eigenvalues that contribute a solution to the CR value is greater than 0.10, then a re-evaluation of
previous matricial expression, the following expres- the pairwise comparisons is recommended. However,
sion is verified: perfect consistency rarely occurs in practice.
Finally, a synthesis must be performed. Synthesis
X
n
If aii ¼ 1; biZ ki ¼ n ð21Þ is the process of weighting and combining priorities
i¼1 throughout the model that leads to the overall results.
Synthesis from the goal node multiplies the weight of
and wi approaches the average of n elements of line i each parent node times the local priorities of its
in the normalized matrix N. children nodes and of those children times the local
If w̄ is calculated from the procedure described in priorities of their children. This process continues
Ref. [28]: down to and including the alternatives.
X
n
a¼ wi ; ð22Þ
i¼1 4. Model for the selection of diagnostic techniques
and instrumentation in a predictive maintenance
and w̄ is replaced by: program
1
w; ð23Þ In the design and planning phase of a PMP, the
a
model for the selection of diagnostic techniques and
is verified [31]: instrumentation in a predictive maintenance program
(MSDT-PMP) can be applied. This decision support
Aw̄ ¼ kmax w̄; ð24Þ system helps to solve an unstructured problem, in
which the decision maker has doubts as to which
where kmax z n.
alternative should be selected.
The closer kmax is to n, the more consistent it is
The decision support model proposed can be
with the comparison matrix A or the more coherent
extended to any other machines or techniques get-
will be the judgments provided. The consistency
ting data from the extension or globalisation phase
index (CI) is used as a measurement of the consisten-
of a PMP. This phase is characterized by the fact
cy of the judgments expressed [28]:
that the time needed to get a return on the invest-
kmax  n ment has been reached and, the number of machines
CI ¼ ð25Þ under control is increased or else new objectives are
n1
set.
Therefore, the CI represents an average of the The PMPs have been categorised at different
eigenvalues. technological levels depending on cost and diagnostic
The consistency ratio (CR) is obtained by dividing capacity [8]:
the CI value by the corresponding random consistency
index (RCI) value as given in Table 3. The RCI was
Level 0. Setup carried out using the control of
evaluated by Saaty through the generation of a ran- sensitive variables. The cost is practically zero and
dom matrix with different dimensions (n) [14,32]. the diagnostic capacity is very low.

Table 3
Values of random consistency index
n 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
RCI 0 0 0.58 0.90 1.12 1.24 1.32 1.41 1.45 1.49 1.51 1.48 1.56 1.57 1.59
 M.C. Carnero / Decision Support Systems 38 (2005) 539–555 547

Level 1. Assumes the use of elementary instru- isolation. This last alternative is the most advanced
mentation, like vibrometers or devices to do the step of technological maintenance.
crackle test. The differentiation between portable and on-line

Level 2. Uses more sophisticated instrumentation, systems is related to the technological levels of a
like vibration analyzers, data processing software, PMP. Therefore, in the case of on-line systems, only
or viscometers. the technological level 3 is considered, corresponding

Level 3. The cost is high as sophisticated analysis to the most technologically evolved.
machines are in use; diagnostic capacity is With regards to the integration of the diagnostic
excellent. techniques, the technological level applied in both
predictive techniques should be similar.
The model is elaborated taking into consideration The procedure developed to elaborate the model
the previous technological levels of the predictive consists of carrying out a factor analysis (FA) with the
techniques, lubricant and vibration analyses and the information supplied by the diagnostic parameters. By
integration of both techniques. doing this, the aim is to eliminate the redundant
When a solution has been obtained from the information and to keep the most relevant information
evaluation of the viability of the setting up of the for a later analysis. When several predictive techni-
PMP, the most appropriate diagnostic technique must ques are applied, the factor analysis also allows the
be selected according to the type of machinery, obtaining of the relevant parameters that favour the
technical and economic characteristics and aspects integration of techniques. The application of this
related to the human resources required, etc. For this technique is due to the fact that the number of
purpose, the model for the selection of diagnostic parameters used in predictive maintenance is high
techniques and instrumentation in a predictive main- and does not always provide information or this
tenance program (MSDT-PMP) has been designed. information is redundant [26]. By using FA, we aim
The selection of lubricant and vibration analyses to get a set of variables that constitutes a coherent
from the range of predictive techniques is due to the subgroup and with independent elements.
fact that these are applied in a higher number of FA is applied to the diagnostic predictive techni-
industrial plants [9]. The introduction of the integra- ques that provide quantitative data. The quantitative
tion of both techniques is due to the fact that the information supplied by the diagnostic techniques
results obtained in this diagnosis are different with selected by the factors resulting from FA is completed
respect to the application of the same techniques in with the incorporation of qualitative information com-

Fig. 2. Hierarchy of MSTD-PMP.

548 M.C. Carnero / Decision Support Systems 38 (2005) 539–555

Table 4 negative influences on the human resources due to

Pairwise comparison matrix and eigenvectors in a PMP based on its application (HR), its maintainability (M) and
lubricant and vibration analyses
easy use (EU).
Technological level 3
Criteria Eigenvector In the case of a PMP integrating lubricant and
D Q COST SUP vibration analyses, the variable capacity of integration
Lubricant analysis (portable system) (INT) is incorporated to favour the integration process
D 1 2 2 3 0.424 of predictive techniques and to avoid the selection of
Q 1/2 1 1 2 0.227 incompatible techniques.
COST 1/2 1 1 2 0.227
The hierarchy elaborated with the goal, decision
SUP 1/3 1/2 1/2 1 0.122
variables and alternatives is shown in Fig. 2.
Vibration analysis (portable system) As examples, the pairwise matrix and eigenvectors
D 1 2 2 3 0.424 obtained from the criteria corresponding to techno-
Q 1/2 1 1 2 0.227
COST 1/2 1 1 2 0.227
Table 5
SUP 1/3 1/2 1/2 1 0.122
Pairwise comparison matrix and eigenvectors in a PMP based on the
integration of lubricant and vibration analyses
Vibration analysis (on-line system)
D 1 2 3 3 0.455 Criteria Eigenvector
Q 1/2 1 1 2 0.141 D Q COST SUP INT
COST 1/3 1 1 1 0.263
SUP 1/3 1/2 1 1 0.141 Technological level 0 (portable system)
D 1 2 2 3 1 0.298
Q 1/2 1 1 2 1/2 0.158
ing from other diagnostic techniques, where the COST 1/2 1 1 2 1/2 0.158
SUP 1/3 1/2 1/2 1 1/3 0.089
results of analysis do not give numerical results. The INT 1 2 2 3 1 0.298
previous process is applied to each technological level
of a PMP and that will be reviewed in this paper. Technological level 1 (portable system)
The most significant diagnostic techniques regard- D 1 2 2 3 1 0.298
ing each technological level related to the factors Q 1/2 1 1 2 1/2 0.158
COST 1/2 1 1 2 1/2 0.158
obtained from the FA and the qualitative diagnostic SUP 1/3 1/2 1/2 1 1/3 0.089
techniques are incorporated as alternatives in a hier- INT 1 2 2 3 1 0.298
archy to which AHP is applied. By means of this
procedure, the techniques that provide redundant Technological level 2 (portable system)
information and the techniques that do not give D 1 2 3 4 1 0.320
Q 1/2 1 1 3 1/3 0.159
relevant information for diagnosis are eliminated and COST 1/3 1 1 3 1/2 0.138
the integration of vibration and lubricant analysis is SUP 1/4 1/3 1/3 1 1/4 0.063
favoured. INT 1 3 2 4 1 0.320
The decision variables used to construct the hier-
archy are: Technological level 3 (portable system)
D 1 2 4 6 1 0.340
Q 1/2 1 2 4 1/2 0.180

Diagnostic quality (D). COST 1/4 1/2 1 2 1/4 0.090

Quantity of failures that can be analyzed ( Q). SUP 1/6 1/4 1/2 1 1/6 0.051

Cost of diagnostic technique (COST). This variable INT 1 2 4 6 1 0.340
is decomposed in investment cost (INVC), setup
Technological level 3 (on-line system)
cost (SETC) and maintenance cost (MANC) of the D 1 3 6 6 1 0.365
diagnostic technique. Q 1/3 1 4 4 1/3 0.163

Supportability of the diagnostic technique (SUP). COST 1/6 1/4 1 1 1/6 0.053
This variable includes: quantity of training needed SUP 1/6 1/4 1 1 1/6 0.053
to apply the technique (T), its portability ( P), INT 1 3 6 6 1 0.365
 M.C. Carnero / Decision Support Systems 38 (2005) 539–555 549

logical level 3 in a PMP based on lubricant analysis can be demonstrated, by means of vibration
and vibration analysis are in Table 4 As can be analysis.
appreciated in Table 4, on-line system is associated
to vibration analysis, because the instrumentation in The acquisition of data used in the analysis is
lubricant analysis is placed in laboratories and the data rather complicated due to the specificity of this kind
are always periodic. of data in the industrial plant, the high cost of
When a PMP is applied based on the integration of acquisition and the restricted access to the data.
lubricant and vibration analyses, the pairwise matrix Nevertheless, two full years of monthly acquisition
and eigenvectors obtained from the criteria corres- of data were developed.
ponding to the different technological levels are in The model selects a minimum of two alternatives
Table 5. between which integration can take place. The capac-
The diagnostic techniques to be applied are depen- ity of integration between lubricant and vibration
dent on the type of machinery, and therefore the techniques and between diagnostic techniques belong-
results shown have been achieved by applying the ing to vibration analysis or lubricant analysis in
MSDT-PMP to screw compressors. isolation has been maintained.
The diagnostic techniques analyzed are in Table 6.
The results of applying FA to diagnostic param-
5. Case study of a screw compressor with the eters in a screw compressor when integration be-
integration of lubricant and vibration analyses tween lubricant and vibration analyses is applied is
described.
A PMP was designed and set up in a petrochem-
ical plant. The program was applied to three screw
compressors. Table 6
The industrial equipment submitted to the analysis Diagnostic techniques analyzed
was adapted for the incorporation of an integrated Diagnostic techniques Code
PMP of vibration and lubricant analysis for the Lubricant analysis
following reasons [7]: Water content in lubricant WACONT
Colour of lubricant COLOUR

Density DENSITY
The equipment has high criticity; its breakdown Content in wear metals (iron) WEAR1
supposes the temporary closedown of the whole Viscosity index VISCINDEX
plant of lubricant production. Content in wear metals (lead) WEAR2

The equipment is rotary, and therefore, adapted for Content in contamination metals (silicon) SI
the application of a PMP based on vibration Total acid number TAN
Viscosity to 100 jC VISC (100)
analysis. Viscosity to 40 jC VISC (40)

Most of the mechanical components of these
compressors are bathed by the same lubricant, Vibration analysis
therefore this can gather information about the Tendency of global vibration value of RMS TEN1
condition of all of them. (10 – 1000 Hz)

Spectral analysis/density of spectral power ES
The compressors are placed in a petrochemical Waterfalls WA
plant and within the area of influence of the plants Spike energy SP
there are two thermal plants. This factor suggests Harmonic tendencies/peak values TEN2
the possibility of the influence of environmental Time signal analysis/form and crest factors TEM
pollution, generating phenomena of grazing and Statistical analysis (kurtosis, variance analysis) KV
Bode diagram BO
corrosion. Polar diagram PO

It is possible to trace the development of Orbital analysis OR
deterioration in the machinery from its initial Finite modal element/experimental FEM
stage, by means of lubricant analysis, up to modal analysis
the stage at which the mechanical damage Cepstrum/envelope CE + EN
550 M.C. Carnero / Decision Support Systems 38 (2005) 539–555

Table 7
Correlation matrix between lubricant and vibration parameters in a screw compressor
WACONT COLOUR DENSITY WEAR1 VISCINDEX TEN1 WEAR2 SI TAN VISC (100) VISC (40)
WACONT 1.00000
COLOUR 0.55024 1.00000
DENSITY 0.36161 0.97058 1.00000
WEAR1 0.16166 0.51704 0.63661 1.00000
VISCINDEX 0.87632 0.28724 0.04897  0.33333 1.00000
TEN1 0.85443 0.27343 0.03393  0.37314 0.99889 1.00000
WEAR2  0.53558 0.17235 0.24485  0.33333  0.33333  0.29639 1.00000
SI 0.38984 0.98306 0.98786 0.51011 0.13912 0.12986 0.32462 1.00000
TAN  0.67486 0.05620 0.29491 0.60150  0.93486  0.94334 0.31162 0.19257 1.00000
VISC (100) 0.40582 0.98133 0.99873 0.62014 0.09914 0.08393 0.22356 0.99067 0.24712 1.00000
VISC (40) 0.18922 0.89371 0.97500 0.73849  0.16644  0.18363 0.26017 0.93635 0.49708 0.96320 1.00000

The correlation matrix between the quantita- tion due to the fact that it has the highest contribution
tive diagnostic techniques analyzed is shown in in variables such as silicon, content colour, density,
Table 7. etc, which are indicative of a contamination process in
The determinant of the correlation matrix is low. the compressor. Factor 2 is called degradation due to
Consequently, there are high intercorrelations between its having the highest contribution of the variables
the variables. This characteristic is necessary in order total acid number or water content which are indica-
to apply factor analysis. tive of a degradation process in the lubricant with a
Due to the quantity of factors available being too lack of additives. Factor 3 is called wear because it
high, a factor analysis has been applied, to obtain a set brings together the two variables that analyzed the
of variables that form a coherent, independent group. wear process in the compressor such as the lead and
Three factors get 100% of the accumulated percentage iron content.
of variance, as can be appreciated in Table 8. As a The diagnostic techniques that provide more infor-
result, only the factors with eigenvalue superior to 1 mation about the contamination, degradation and wear
are preserved (Kaiser rule). process (results of factor analysis) in the compressor
The rotation through varimax simplifies the results are selected as alternatives. These alternatives are
(Table 9) and facilitates interpretation of the data. As introduced in the hierarchy of Fig. 2 joint with the
can be appreciated in Table 9, each variable is only alternatives that give qualitative information. So, the
saturated in one factor and each factor has distinct alternatives considered by technological level are in
load distribution. Thus factor 1 is called contamina-

Table 8
Integration of diagnostic parameters in factors Table 9
Results provided before applying a rotation through varimax
Factor Eigenvalue Percentage of Accumulated
variance percentage of Factors
variance Contamination Degradation Wear
1 5.57269 50.7 50.7 WATCONT 0.37138 0.83074 0.41466
2 4.02062 36.6 87.2 COLOUR 0.97404 0.22638 0.00107
3 1.40669 12.8 100.0 DENSITY 0.99987  0.01477 0.00615
4 0.00000 0.0 100.0 WEAR1 0.62651  0.41484 0.65985
5 0.00000 0.0 100.0 VISCINDEX 0.06332 0.99613 0.06097
6 0.00000 0.0 100.0 TEN1 0.04859 0.99869 0.01623
7 0.00000 0.0 100.0 WEAR2 0.24623  0.29375  0.92363
8 0.00000 0.0 100.0 SI 0.98992 0.08373  0.11419
9 0.00000 0.0 100.0 TAN 0.28052  0.95891 0.04237
10 0.00000 0.0 100.0 VISC (100) 0.99929 0.03521 0.01315
11 0.00000 0.0 100.0 VISC (40) 0.97139  0.23196 0.05105
 M.C. Carnero / Decision Support Systems 38 (2005) 539–555 551

Table 10
Diagnostic techniques in each technological level in a PMP based on integrating vibration and lubricant analysis
Portable system On-line system
Level 0 Level 1 Level 2 Level 3 Level 3
Content in wear and Content in wear and Content in wear and Content in wear and Content in wear and
contamination contamination metals contamination metals contamination metals contamination metals
metals Spectral analysis Spectral analysis/density Spectral analysis/density
of spectral power of spectral power
Vibration analysis Tendency of global Viscosity to 40 jC Viscosity to 40 and 100 jC Viscosity to 40 and 100 jC
vibration value RMS
(10 – 1000 Hz)
Colour of lubricant Viscosity to 40 jC Waterfalls Waterfalls Waterfalls
Tendency of global Tendency of global Tendency of global
vibration value RMS vibration value RMS vibration value RMS
(10 – 1000 Hz) (10 – 1000 Hz) (10 – 1000 Hz)
Spike energy Spike energy Spike energy
Water content Water content Harmonic tendencies/ Cepstrum/envelope Cepstrum/envelope
peak values
Water content Water content/total Water content/total
acid number acid number
Time signal analysis/form Time signal analysis/form Polar diagram
and crest factors and crest factors

Table 10. Each of these alternatives has associated a has values which are inferior to 0.1 in all the cases,
particular instrumentation in agreement with their and therefore, is considered acceptable.
technological level.
Next, the AHP is applied. 6.1. Portable system
The maximum number of alternatives permitted in
AHP is nine, and therefore, this is the number of 6.1.1. Technological level 0 (Table 11)
alternatives or diagnostic techniques considered in the The diagnostic techniques analyzed provide infor-
technological levels 2 and 3 of the model. Although mation about the degradation in colour, water content
there are other diagnostic techniques, the more repre- in the inspection of free water, contamination by
sentatives have been included. particles and preferably wear and anomalous mechan-
The diagnostic techniques have been adapted to ical behaviour. The complementary nature of each
each technological level. Thus, water content is in- technique is demonstrated by the close preferences.
cluded in level 0 by means of visual inspection and in The existence of similar values recommends the
level 2 by means of a Karl Fischer device. Therefore, application of all the parameters to the industrial plant,
the results include the instrumentation needed to apply and this is beneficial when a total productive mainte-
each of the diagnostic techniques in each technolog- nance is combined with a PMP.
ical level. The model selects visual inspection of particles in
lubricant and visual inspection of vibration (or use of
screwdriver), favouring the integration process of
6. Results lubricant and vibration analyses.

In this section, the results of the model after 6.1.2. Technological level 1
applying factor analysis to the data obtained in screw Table 11 shows that the global preferences of
compressors of a petrochemical plant, and by inte- alternatives are very close, and therefore the use of
grating qualitative and quantitative variables from all the techniques applying the concept of comple-
lubricant and vibration analyses are presented. As mentarity is recommended, unless the plant is inter-
can be appreciated in Table 11, the consistency ratio ested in a limited number of techniques, in which case
552 M.C. Carnero / Decision Support Systems 38 (2005) 539–555

Table 11 Table 11 (continued)

Hierarchy of the diagnostic techniques in technological levels 0, 1, 2 Selection of diagnostic techniques in a PMP based on integrated
and 3 lubricant and vibration analyses
Selection of diagnostic techniques in a PMP based on integrated
Diagnostic technique Instrumentation Preferences
lubricant and vibration analyses
Spike energy IRD spectral analyzer 0.071
Diagnostic technique Instrumentation Preferences
Tendency of global Vibrometer/spectral 0.060
Technological level 0 vibration value RMS analyzer
Content in wear and Visual inspection of 0.275 (10 – 1000 Hz)
contamination metals particles in lubricant Time signal Spectral analyzer/ 0.052
(no instrumentation) analysis/form oscilloscope
Vibration analysis Visual inspection/use 0.275 and crest factors
of screwdriver Consistency ratio = 0.00
Colour of lubricant Visual inspection 0.225
(No instrumentation)
Water content Visual inspection 0.225
(No instrumentation) viscosity and tendencies of global vibration value
Consistency ratio = 0.00 RMS between 10 and 1000 Hz can be applied. The
use of the stain of oil technique can be considered as a
Technological level 1 support technique.
Viscosity to 40 jC Capillary viscometer 0.305
Tendency of global Vibrometer 0.262
vibration value RMS 6.1.3. Technological level 2
(10 – 1000 Hz) The model supplies (see Table 11) diagnostic
Content in wear and Stain of oil 0.229 techniques with quality of diagnosis and capacity
contamination metals for integration with other techniques. The model
Water content Crackle test 0.204
suggested the application of spectral analysis and
Consistency ratio = 0.00
waterfalls, particle counting and viscosity control.
Technological level 2 Therefore, the instrumentation required is: spectral
Spectral analysis Spectral analyzer 0.200 analyzer, particle meter and capillary viscometer.
Waterfalls Spectral analyzer 0.180 The other alternatives have preferences inferior to
Content in wear and Particle meter 0.172
the aforementioned. It should be pointed out that the
contamination metals
Viscosity to 40 jC Capillary viscometer 0.120 tendency of global vibration value of RMS between
Tendency of global Vibrometer/ 0.087 10 and 1000 Hz owes its classification to the lower
vibration value rms spectral analyzer setup and maintenance cost, but considering that the
(10 – 1000 Hz) industrial plant has a high technological level, the
Spike energy IRD spectral analyzer 0.077
cost variable should not influence the selection of
Harmonic tendencies/ Spectral analyzer 0.068
peak values alternatives.
Water content Karl Fischer 0.053
Time signal analysis/form Spectral analyzer/ 0.044 6.1.4. Technological level 3
and crest factors oscilloscope The classification of alternatives shown in Table
Consistency ratio = 0.04
11 suggests the application of more technological
Technological level 3 techniques to provide better quality in the diagnosis
Content in wear and Spectrometer of 0.224 and a superior capacity for the protection of machin-
contamination metals atomic absorption ery, because they detect the failures and the deterio-
Spectral analysis/density Spectral analyzer 0.150 ration of elements and lubricants in early phases of
of spectral power
development. The use of spectrografy, a technique
Viscosity to 40 Automatic viscometer 0.135
and 100 jC that defines the factors that are contributing to wear
Waterfalls Spectral analyzer 0.118 and contamination, together with spectral analysis
Water content/total Karl Discher/tritrador 0.109 that detects wear effects are the techniques that
acid number supply the most reliable and fast diagnoses. These
Cepstrum/envelope Advanced oscilloscope 0.080
diagnostic techniques also provide information about
 M.C. Carnero / Decision Support Systems 38 (2005) 539–555 553

Table 12 mented with viscosity and waterfalls, as the model

Hierarchy of the diagnostic techniques in technological level 3 in a suggests.
PMP based on continuous integrated lubricant and vibration
analyses The model minimizes the repetitive information in
the process of selection of diagnostic techniques.
Selection of diagnostic techniques in a PMP based on integrated
lubricant and vibration analyses
6.2. On-line system
Diagnostic technique Instrumentation Preferences
Content in wear and Spectrometer of 0.222
6.2.1. Technological level 3
contamination metals atomic absorption
Spectral analysis/density Spectral analyzer 0.141 The procedure used is similar to that in techno-
of spectral power logical level 3 in a portable system, although it
Viscosity to 40 and 100 jC Automatic viscometer 0.131 rejects the alternatives with the lower preferences
Waterfalls Spectral analyzer 0.121 obtained in the level 3 of a PMP based on vibration
Water content/total Karl Fischer/tritrador 0.107
analysis. This means the analysis can be limited to
acid number
Polar diagram Continuous acquisition 0.079 nine alternatives. As can be seen in Table 12, the
system/displacement classification is very similar to that obtained in a
sensors/key phasor portable system, although the technology applied is
Cepstrum/envelope Advanced oscilloscope 0.079 superior in this case, because it provides a more
Spike energy IRD spectral analyzer 0.065
exact control of the state of machinery. The quantity
Tendency of global Vibrometer/spectral 0.055
vibration value RMS analyzer of techniques applied in the plant depends on eco-
(10 – 1000 Hz) nomic variables and the criticality of industrial
machinery, which is generally very elevated in this
Consistency ratio = 0.02 technological level. Therefore, the use of the follow-
ing instrumentation is suggested: spectrometer of
the most suitable moment in which to carry out the atomic absorption, spectral analyzer and automatic
change of lubricant due to contamination or loss of viscometer. This allows the application of diagnostic
protection capacity. These techniques can be comple- techniques: content in wear and contamination met-

Fig. 3. Sensitivity analysis corresponding to technological level 3 in on-line system.

554 M.C. Carnero / Decision Support Systems 38 (2005) 539–555

als, spectral analysis and density of spectral power, tenance model for two-unit series system, European Journal of
viscosity to 40 and 100 jC and waterfalls. The Operational Research 116 (1999) 281 – 290.
[3] P. Beltrán, A. López, El Mantenimiento Predictivo en aeroge-
greater weight given to the polar diagram rather than neradores. Caso práctico: estudio de averı́as, Proceedings 4j-
the cepstrum/envelope is due to the capacity of the Congreso Español de Mantenimiento, AEM, Barcelona, 2000.
first to provide information about the behaviour of [4] J.E. Berry, Good Vibes About Oil Analysis, Practicing Oil
axis, an aspect that cannot be analyzed with any Analysis, J. Fitch, Tulsa, 1999 (November – December).
[5] B.K.N. Rao, Handbook of Condition monitoring, Elsevier,
other alternatives.
Oxford, 1996.
The sensitive analysis corresponding to the setup [6] L. Borao, M. Garcı́a, Mantenimiento Predictivo. Implantación
of a PMP based on integrating lubricant and vibration Industrial. Implantación de un programa predictivo, Manteni-
analyses provides stable results in all the technolog- miento, no. 97, 1996 (Septiembre) 13 – 17.
ical levels. Fig. 3 shows an example of the sensitivity [7] M.C. Carnero, Evaluación del ciclo de vida de un Programa de
analysis corresponding to technological level 3 in an Mantenimiento Predictivo mediante técnicas multicriterio,
Thesis, University of Castilla-La Mancha, ETSII, Ciudad
on-line system. Real, 2001.
[8] M.C. Carnero, E. La Torre, M.A. Alcázar, J. Conde, Control
of wear applied to compressor: trends in lubricant analysis,
7. Conclusions International Journal on the Science and Technology of Fric-
tion Lubrication and Wear 225 – 229 (1999) 905 – 912.
[9] A.H. Christer, W. Wang, J.M. Sharp, A state space condition
In the decision support model designed, technolog- monitoring model for furnace erosion prediction and replace-
ical and organizational issues have been incorporated ment, European Journal of Operational Research 101 (1997)
that until now had not been sufficiently researched in 1 – 14.
the topic of a predictive maintenance program. [10] Computational Systems Inc., PC-based integration of spectro-
graphic, Ferrographic and Vibration analysis data, P/PM Tech-
Vibration analysis and lubricant analysis are the
nology, 1991 (January – February).
most frequently applied predictive techniques at pres- [11] M. Cuesta, F.J. Herrero, Introducción al Análisis Factorial,
ent, as a result of which the integration of both Tutorial:DPAM#95.2, Oviedo University, 2002.
techniques in a single predictive maintenance program [12] D.J. Edwards, G.D. Holt, F.C. Harris, Predictive maintenance
can provide significant benefits for the company. techniques and their relevance to construction plant, Journal of
A model of selection of diagnostic techniques and Quality in Maintenance Engineering 4 (1) (1998) 25 – 37.
[13] J.C. Fitch, Proactive and Predictic Strategies for Setting
instrumentation in a predictive maintenance program Alarms and Limits of Oil Analysis, Noria, Tulsa, 1998.
(MSDT-PMP) has been developed. Factor analysis [14] E. Forman, M.A. Selly, Decisión by Objetives, (World Scien-
and AHP have been combined. The model is applied tific, London, 2001).
to different technological levels in PMPs based on [15] A.K.S. Jardine, V. Makis, D. Banjevic, D. Braticevic, M.
integrated lubricant and vibration in screw compres- Ennis, A decision optimization model for condition-based
maintenance, Journal of Quality in Maintenance Engineering
sors placed in a petrochemical plant. 4 (2) (1998) 115 – 121.
The results obtained will facilitate the decision [16] A.K.S. Jardine, T. Joseph, D. Banjevic, Optimizing condition-
making of the planner of the predictive maintenance based maintenance decisions for equipment subject to vibration
program, as well as favour the development of the monitoring, Journal of Quality in Maintenance Engineering 5
integration of predictive techniques, an aspect that (3) (1999) 192 – 202.
[17] B. Johnson, Oil Analysis Success at A Power Generation
currently lacks models for making decisions, due to Station, Practicing Oil Analysis, J. Fitch, Tulsa, 1998 (July –
the technical and organizational difficulties that its August).
application represents, aspects in which this article [18] V. Kakkar, Ontario power generation’s nanticoke power plant
aims to contribute. vol. 20, no. 4, Orbit, Bently, NV, 1999.
[19] G.M. Knapp, R. Javadpour, H. Wang, An ARTMAP neural
networkbased machine condition monitoring system, Journal
of Quality in Maintenance Engineering 6 (2) (2000) 86 – 105.
References [20] T. Lund-Hansen, Innovate condition monitoring methodologies
for improved plant economics, Proceedings del Sixteenth An-
[1] G. Barbara, S. Tabachnick, S. Linda, Using Multivariate Sta- nual Meeting and Seminar of Canadian Machinery Vibration
tistic, HarperColling Publishing, New York, 1983. Association (CMVA), Toronto, Canadá, 1997 (November).
[2] F. Barbera, H. Schneider, E. Watson, A condition based main- [21] M. Lupinucci, J.G. Pérez Davila, L. Tiseyra, Improving sheet
 M.C. Carnero / Decision Support Systems 38 (2005) 539–555 555

metal quality and producto throughput with bently’s machinery [34] E. Triantaphyllou, B. Kovalerchuck, L.J.R. Mann, J. Knapp,
management system vol. 21, no. 3, Orbit, Bently, NV, 2000. Determining the most important criteria in maintenance deci-
[22] K. Mobley, Why predictive programs fail, Plant Services, sion making, Journal of Quality in Maintenance Engineering 3
1997 (October). (1) (1997) 16 – 28.
[23] K. Mobley, Predictive maintenance equipment, The 1998 [35] D.D. Troyer, Let’s Integrate Oil Analysis and Vibration Anal-
CMMS, PM/PdM Handbook, Putman publishing, Itasca, IL, ysis, Practicing Oil Analysis, J. Fitch, Tulsa, 1998 (July –
1998. August).
[24] Nasa, Appendix H. Predictive Testing and Inspection, Work- [36] A.H.C. Tsang, Strategic dimensions of maintenance manage-
ing paper, Nasa handbook, Octubre, 1994. ment, Journal of Quality in Maintenance Engineering 8 (1)
[25] B. Johnson, Oil analysis success at a Power Generation Sta- (2002) 7 – 39.
tion, Practicing Oil Analysis, J. Fitch, Tulsa, 1998 (July – [37] A. Valverde, Análisis de la disponibilidad de los equipos di-
August). námicos y su incidencia en el mantenimiento en plantas in-
[26] Preditec, Curso de Introducción al Análisis Predictivo de Lu- dustriales, Thesis, UNED, 1994.
bricantes, Zaragoza, Julio, 1997. [38] J.M. Villar, L.O. Masson, J.A. Gomes, Proactive mainte-
[27] C.E. Reese, C.H. Lochmüller, Introduction to Factor Analysis, nance—a successful history vol. 21 no. 3, Orbit, Bently,
Duke University, Durham, 1994. NV, 2000.
[28] T.L. Saaty, The Analytic Hierarchy Process, McGraw Hill, [39] Wearcheck. http://www.wearcheck.com/info/about_interpreta-
New York, 1980. tion.asp, (2003).
[29] H. Saranga, Relevant condition-parameter strategy for an ef- [40] J.W. Weyerhaeuser, Bearing Failures Dry Up at Weyerhaeuser,
fective condition-based maintenance, Journal of Quality in Practicing Oil Analysis, J. Fitch, Tulsa, 2000 (March – April).
Maintenance Engineering 8 (1) (2002) 92 – 105.
[30] D.J. Sherwin, B. Al-Najjar, Practical models for condition
monitoring inspection intervals, Journal of Quality in Main- Ma. C. Carnero Moya received her PhD from the University of
tenance Engineering 5 (3) (1999) 203 – 220. Castilla-La Mancha. Her research interests are in decision support
[31] H.A. Taha, Investigación de operaciones, Una introducción, systems, multiple criteria decision making, evaluation system of
Pearson, México, 1998. maintenance policies and in the theories and applications of condi-
[32] E. Triantaphyllou, S.H. Mann, Using the analytic hierarchy tion based maintenance. She has published in different journals
process for decision making In engineering applications: some including International Journal of Lubrication and Wear and Quality
challenges, International Journal of Industrial Engineering: Progress. She is a professor in the Technical School of Industrial
Applications and Practice 2 (1) (1995) 35 – 44. Engineering (University of Castilla-La Mancha), and has partici-
[33] E. Triantaphyllou, F.A. Lootsma, P.M. Pardalos, S.H. Mann, pated in some project about Condition Based Maintenance, sup-
On the evaluation and application of different scales for quan- ported by the European Union and Regional Administration.
tifying pairwise comparisons in Fuzzy Sets, Journal of Multi-
Criteria Decision Analysis 3 (3) (1994) 133 – 155.