Beruflich Dokumente
Kultur Dokumente
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
The continuous production process requires a high
degree of availability and the elimination of unexpected breakdown that could cause a prolonged stoppage in production [9]. Predictive maintenance can
contribute to improving plant availability, safety,
quality, reduction of maintenance costs, etc. This
has led to an increase in the number of predictive
maintenance programs (PMPs) applied, but, during
the setting up of a PMP, there is a number of decisions
involved that lack decision support systems or mod-
540
541
542
Table 1
Diagnostic techniques in lubricant and vibration analyses [8]
Vibration analysis
Lubricant analysis
Spectral analysis
Viscosity
Index of viscosity
Water content
Total acid number
Total basic number
Analysis of harmonic
and orders
Trend of the global
value of vibration
Cepstrum
Analysis of
temporary signal
Spike energy
Bode plot
Insoluble in pentane
and benzene
Freezing point
Igniting point
Combustion point
Demulsibility
Tendency to foaming
Tendency to formation of
coal and ash content
Polar plot
Waterfalls
Orbital analysis
Statistical analysis
Hilbert transform
Envelope
Modal analysis
Each of the predictive techniques detects deterioration in different phases of its evolution [25].
Lubricant analysis is capable of detecting the
anomaly in the early phases of its development,
whereas vibration analysis will only be able to
evaluate it when the breakdown has already
occurred.
Detection of the root causes of the failures [24,35].
Consequently, there is an intensification and improvement of the level of information regarding
incidents that can be transmitted to the personnel in
charge of corrective activities.
Nevertheless, the simultaneous application of both
predictive techniques does not necessarily mean their
integration. The computer applications that have been
reviewed have designed the databases of each of the
predictive technologies on an independent basis and
without establishing any link of union between them
[23]. Furthermore, the instrumentation used in each of
the predictive techniques usually comes from different
manufacturers, which also impedes the integration of
the information [22].
The capacity of integration of particle counting and
ferrography techniques with spectral analysis is demonstrated in the processes of wear and pollution that
lead to a mechanical breakdown [10]. The condition
of the equipment can be determined by means of
spectral analysis, and ferrographical analysis will
determine with accuracy the element or component
that is developing the anomaly; particle counting can
also determine the gradient of change of the condition
and determine if the initial cause of the failure is due
to an external factor. The periodic use of spectra in
waterfall is suggested as a diagnostic technology, as it
is a technique that identifies the condition of the
equipment and the failure that has developed, aspects
that cannot be investigated by means of other diagnostic techniques; moreover, as it is carried out by
means of comparison, it restricts the time of identification of the failure.
The application of the diagnostic techniques on an
independent basis is recommended until as much
knowledge as possible has been obtained using each
technique individually, in order to later proceed to
their integration. Once this level has been reached, it
would be convenient to have software available that
incorporates utilities for this integration.
543
p
If V L is called A (factor loading matrix), then
R AAV
Aunrorated Arotated
Aunrotated Aunrotated
R
V
L VVRV
Rres R R
R VLVV
10
544
11
13
i; j 1; 2; . . . ; n;
bi; j; k:
15
should always be true for any combination of comparisons taken from the judgment matrix.
16
i; j 1; 2; . . . ; n;
17
and then:
n
X
aij
j1
or:
n
X
wj
n i 1; 2; . . . ; n;
wi
aij wj nwi
i 1; 2; . . . ; n;
18
19
j1
545
20
Table 2
Scale of relative importances [28]
Intensity of
importance
Verbal scale
Explanation
1
3
Equal importance
Weak importance of one over another
Demonstrated importance
Absolute importance
2, 4, 6, 8
Reciprocals of
above numbers
546
biZ
n
X
ki n
21
i1
n
X
22
wi ;
i1
23
is verified [31]:
Aw kmax w;
24
where kmax z n.
The closer kmax is to n, the more consistent it is
with the comparison matrix A or the more coherent
will be the judgments provided. The consistency
index (CI) is used as a measurement of the consistency of the judgments expressed [28]:
CI
In the AHP, the pairwise comparisons in a judgment matrix are considered to be adequately consistent if the corresponding CR is less than 10%. If the
CR value is greater than 0.10, then a re-evaluation of
the pairwise comparisons is recommended. However,
perfect consistency rarely occurs in practice.
Finally, a synthesis must be performed. Synthesis
is the process of weighting and combining priorities
throughout the model that leads to the overall results.
Synthesis from the goal node multiplies the weight of
each parent node times the local priorities of its
children nodes and of those children times the local
priorities of their children. This process continues
down to and including the alternatives.
kmax n
n1
25
Table 3
Values of random consistency index
n
RCI
1
0
2
0
3
0.58
4
0.90
5
1.12
6
1.24
7
1.32
8
1.41
9
1.45
10
1.49
11
1.51
12
1.48
13
1.56
14
1.57
15
1.59
Level 1. Assumes the use of elementary instrumentation, like vibrometers or devices to do the
crackle test.
Level 2. Uses more sophisticated instrumentation,
like vibration analyzers, data processing software,
or viscometers.
Level 3. The cost is high as sophisticated analysis
machines are in use; diagnostic capacity is
excellent.
The model is elaborated taking into consideration
the previous technological levels of the predictive
techniques, lubricant and vibration analyses and the
integration of both techniques.
When a solution has been obtained from the
evaluation of the viability of the setting up of the
PMP, the most appropriate diagnostic technique must
be selected according to the type of machinery,
technical and economic characteristics and aspects
related to the human resources required, etc. For this
purpose, the model for the selection of diagnostic
techniques and instrumentation in a predictive maintenance program (MSDT-PMP) has been designed.
The selection of lubricant and vibration analyses
from the range of predictive techniques is due to the
fact that these are applied in a higher number of
industrial plants [9]. The introduction of the integration of both techniques is due to the fact that the
results obtained in this diagnosis are different with
respect to the application of the same techniques in
547
548
Table 4
Pairwise comparison matrix and eigenvectors in a PMP based on
lubricant and vibration analyses
Technological level 3
Criteria
D
Eigenvector
Q
COST
SUP
3
2
2
1
0.424
0.227
0.227
0.122
3
2
2
1
0.424
0.227
0.227
0.122
3
2
1
1
0.455
0.141
0.263
0.141
Eigenvector
Q
COST
SUP
INT
1
1/2
1/2
1/3
1
0.298
0.158
0.158
0.089
0.298
1
1/2
1/2
1/3
1
0.298
0.158
0.158
0.089
0.298
1
1/3
1/2
1/4
1
0.320
0.159
0.138
0.063
0.320
1
1/2
1/4
1/6
1
0.340
0.180
0.090
0.051
0.340
1
1/3
1/6
1/6
1
0.365
0.163
0.053
0.053
0.365
549
Table 6
Diagnostic techniques analyzed
Diagnostic techniques
Code
Lubricant analysis
Water content in lubricant
Colour of lubricant
Density
Content in wear metals (iron)
Viscosity index
Content in wear metals (lead)
Content in contamination metals (silicon)
Total acid number
Viscosity to 100 jC
Viscosity to 40 jC
WACONT
COLOUR
DENSITY
WEAR1
VISCINDEX
WEAR2
SI
TAN
VISC (100)
VISC (40)
Vibration analysis
Tendency of global vibration value of RMS
(10 1000 Hz)
Spectral analysis/density of spectral power
Waterfalls
Spike energy
Harmonic tendencies/peak values
Time signal analysis/form and crest factors
Statistical analysis (kurtosis, variance analysis)
Bode diagram
Polar diagram
Orbital analysis
Finite modal element/experimental
modal analysis
Cepstrum/envelope
TEN1
ES
WA
SP
TEN2
TEM
KV
BO
PO
OR
FEM
CE + EN
550
Table 7
Correlation matrix between lubricant and vibration parameters in a screw compressor
WACONT COLOUR DENSITY WEAR1
WACONT
1.00000
COLOUR
0.55024 1.00000
DENSITY
0.36161 0.97058
WEAR1
0.16166 0.51704
VISCINDEX
0.87632 0.28724
TEN1
0.85443 0.27343
WEAR2
0.53558 0.17235
SI
0.38984 0.98306
TAN
0.67486 0.05620
VISC (100)
0.40582 0.98133
VISC (40)
0.18922 0.89371
1.00000
0.63661
0.04897
0.03393
0.24485
0.98786
0.29491
0.99873
0.97500
VISCINDEX TEN1
1.00000
0.33333
1.00000
0.37314
0.99889
0.33333 0.33333
0.51011
0.13912
0.60150 0.93486
0.62014
0.09914
0.73849 0.16644
The correlation matrix between the quantitative diagnostic techniques analyzed is shown in
Table 7.
The determinant of the correlation matrix is low.
Consequently, there are high intercorrelations between
the variables. This characteristic is necessary in order
to apply factor analysis.
Due to the quantity of factors available being too
high, a factor analysis has been applied, to obtain a set
of variables that form a coherent, independent group.
Three factors get 100% of the accumulated percentage
of variance, as can be appreciated in Table 8. As a
result, only the factors with eigenvalue superior to 1
are preserved (Kaiser rule).
The rotation through varimax simplifies the results
(Table 9) and facilitates interpretation of the data. As
can be appreciated in Table 9, each variable is only
saturated in one factor and each factor has distinct
load distribution. Thus factor 1 is called contaminaTable 8
Integration of diagnostic parameters in factors
1.00000
0.29639
0.12986
0.94334
0.08393
0.18363
WEAR2 SI
1.00000
0.32462
0.31162
0.22356
0.26017
TAN
1.00000
0.19257 1.00000
0.99067 0.24712 1.00000
0.93635 0.49708 0.96320
1.00000
Table 9
Results provided before applying a rotation through varimax
Factor
Eigenvalue
Percentage of
variance
Accumulated
percentage of
variance
1
2
3
4
5
6
7
8
9
10
11
5.57269
4.02062
1.40669
0.00000
0.00000
0.00000
0.00000
0.00000
0.00000
0.00000
0.00000
50.7
36.6
12.8
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
50.7
87.2
100.0
100.0
100.0
100.0
100.0
100.0
100.0
100.0
100.0
Factors
WATCONT
COLOUR
DENSITY
WEAR1
VISCINDEX
TEN1
WEAR2
SI
TAN
VISC (100)
VISC (40)
Contamination
Degradation
Wear
0.37138
0.97404
0.99987
0.62651
0.06332
0.04859
0.24623
0.98992
0.28052
0.99929
0.97139
0.83074
0.22638
0.01477
0.41484
0.99613
0.99869
0.29375
0.08373
0.95891
0.03521
0.23196
0.41466
0.00107
0.00615
0.65985
0.06097
0.01623
0.92363
0.11419
0.04237
0.01315
0.05105
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
Vibration analysis
Viscosity to 40 jC
Colour of lubricant
Tendency of global
vibration value RMS
(10 1000 Hz)
Viscosity to 40 jC
Water content
Water content
Waterfalls
Tendency of global
vibration value RMS
(10 1000 Hz)
Spike energy
Cepstrum/envelope
Waterfalls
Tendency of global
vibration value RMS
(10 1000 Hz)
Spike energy
Cepstrum/envelope
Water content/total
acid number
Time signal analysis/form
and crest factors
Water content/total
acid number
Polar diagram
Waterfalls
Tendency of global
vibration value RMS
(10 1000 Hz)
Spike energy
Harmonic tendencies/
peak values
Water content
Time signal analysis/form
and crest factors
6. Results
552
Table 11
Hierarchy of the diagnostic techniques in technological levels 0, 1, 2
and 3
Selection of diagnostic techniques in a PMP based on integrated
lubricant and vibration analyses
Diagnostic technique
Technological level 0
Content in wear and
contamination metals
Vibration analysis
Colour of lubricant
Water content
Instrumentation
Preferences
Visual inspection of
particles in lubricant
(no instrumentation)
Visual inspection/use
of screwdriver
Visual inspection
(No instrumentation)
Visual inspection
(No instrumentation)
0.275
Capillary viscometer
Vibrometer
0.305
0.262
Stain of oil
0.229
Crackle test
0.204
Spectral analyzer
Spectral analyzer
Particle meter
0.200
0.180
0.172
Capillary viscometer
Vibrometer/
spectral analyzer
0.120
0.087
0.077
0.068
Karl Fischer
Spectral analyzer/
oscilloscope
0.053
0.044
Spectrometer of
atomic absorption
Spectral analyzer
0.224
0.150
Automatic viscometer
0.135
Spectral analyzer
Karl Discher/tritrador
0.118
0.109
Advanced oscilloscope
0.080
0.275
Technological level 2
Spectral analysis
Waterfalls
Content in wear and
contamination metals
Viscosity to 40 jC
Tendency of global
vibration value rms
(10 1000 Hz)
Spike energy
Harmonic tendencies/
peak values
Water content
Time signal analysis/form
and crest factors
Consistency ratio = 0.04
Technological level 3
Content in wear and
contamination metals
Spectral analysis/density
of spectral power
Viscosity to 40
and 100 jC
Waterfalls
Water content/total
acid number
Cepstrum/envelope
Instrumentation
Preferences
Spike energy
Tendency of global
vibration value RMS
(10 1000 Hz)
Time signal
analysis/form
and crest factors
Consistency ratio = 0.00
0.071
0.060
Spectral analyzer/
oscilloscope
0.052
0.225
0.225
Table 11 (continued)
Instrumentation
Preferences
Spectrometer of
atomic absorption
Spectral analyzer
0.222
0.141
Automatic viscometer
Spectral analyzer
Karl Fischer/tritrador
0.131
0.121
0.107
Continuous acquisition
system/displacement
sensors/key phasor
Advanced oscilloscope
IRD spectral analyzer
Vibrometer/spectral
analyzer
0.079
Cepstrum/envelope
Spike energy
Tendency of global
vibration value RMS
(10 1000 Hz)
0.079
0.065
0.055
553
554
[3]
[4]
[5]
[6]
[7]
[8]
7. Conclusions
In the decision support model designed, technological and organizational issues have been incorporated
that until now had not been sufficiently researched in
the topic of a predictive maintenance program.
Vibration analysis and lubricant analysis are the
most frequently applied predictive techniques at present, as a result of which the integration of both
techniques in a single predictive maintenance program
can provide significant benefits for the company.
A model of selection of diagnostic techniques and
instrumentation in a predictive maintenance program
(MSDT-PMP) has been developed. Factor analysis
and AHP have been combined. The model is applied
to different technological levels in PMPs based on
integrated lubricant and vibration in screw compressors placed in a petrochemical plant.
The results obtained will facilitate the decision
making of the planner of the predictive maintenance
program, as well as favour the development of the
integration of predictive techniques, an aspect that
currently lacks models for making decisions, due to
the technical and organizational difficulties that its
application represents, aspects in which this article
aims to contribute.
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
References
[1] G. Barbara, S. Tabachnick, S. Linda, Using Multivariate Statistic, HarperColling Publishing, New York, 1983.
[2] F. Barbera, H. Schneider, E. Watson, A condition based main-
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
555