Vous êtes sur la page 1sur 15

Condition-based

Condition-based maintenance: maintenance


tools and decision making
Albert H.C. Tsang
The Hong Kong Polytechnic University, Hong Kong 3

Options for preventive maintenance


Historically, maintenance activities have been regarded as a necessary evil by
the various management functions in an organization. However, in recent years
this attitude has increasingly been replaced by one which recognizes
maintenance as a strategic issue in the organization. The developments which
contributed to this change include: environmental concerns, safety issues,
warranty and liability factors, regulatory matters, ageing plant and equipment,
drive for cost reduction and the like. To live up to the new expectations
demanded of maintenance activities, maintenance programmes have to be
developed to ensure that physical assets will continue to fulfil their intended
functions at a minimum expenditure of resources. Obviously, maintenance
activities which do not contribute to preserving or restoring the intended
functions of assets should be eliminated.
Maintenance activities fall into two broad categories, namely corrective
maintenance and preventive maintenance. Corrective maintenance (CM), also
known as breakdown maintenance, is performed when action is taken to restore
the functional capabilities of failed or malfunctioned equipment or systems.
This is a reactive approach to maintenance because the action is triggered by
the unscheduled event of an equipment failure. With this kind of maintenance
policy, the maintenance related costs are usually high due to the following
reasons:
● the high cost of restoring equipment to an operable condition under crisis
situation;
● the secondary damage and safety/health hazards inflicted by the failure;
● the penalty associated with lost production.
Preventive maintenance (PM), on the other hand, is the approach developed to
avoid this kind of waste. Traditionally, PM takes mainly the form of equipment
overhaul or item replacement at fixed intervals. This practice is known as time-
directed (TD) maintenance. When only TD maintenance is performed, however,
a lot of opportunities for enhancement of the PM programme will be lost.

The author would like to thank Professor A.K.S. Jardine for helping to review the manuscript. Journal of Quality in Maintenance
Engineering,
This work was also supported by the Natural Sciences and Engineering Research Council of Vol. 1 No. 3, 1995, pp. 3-17.
Canada under Grant A4174. © MCB University Press, 1355-2511
JQME To exploit the full potential of PM, the reasons for performing PM tasks
1,3 should be recognized. The reasons are[1]:
● to prevent failure;
● to detect the onset of failure;
● to discover hidden failure.
4 These reasons give rise to three types of PM task. The first and most common
type of PM task is time directed (TD). TD tasks, performed to prevent or retard
failures, are done at hard time intervals regardless of other information that
may be available when the preset time occurs. A TD task also requires an
intrusion into the equipment, thereby rendering it out of service until the task is
completed.
The second type of PM task is condition-based maintenance (CBM), which is
also known as on-condition maintenance, condition-directed maintenance, or
predictive maintenance. CBM is designed to detect the onset of a failure. It is an
appropriate option for PM when the following conditions apply:
● either failure prevention is not feasible, or how it can be achieved is not
yet known, as in cases where the event leading to failure occurs in a
predominantly random manner;
● a measurable parameter which correlates with the onset of failure has
been identified, for example, the solids content in the lubricant is an
indicator of the machine’s wearing condition;
● it is possible to identify a value of that parameter when action may be
taken before full failure occurs, such as the setting of warning limits for
the solids content of the lubricant.
CBM is similar to TD maintenance in that the task is performed at preset
intervals. However, unlike TD tasks, CBM does not normally involve an
intrusion into the equipment and the actual preventive action is taken only
when it is believed that an incipient failure has been detected.
Standby units, protective devices, or infrequently used equipment create
special problems in preventive maintenance. Failures in this type of equipment
are known as hidden failures because they are not evident until the time when
the proper function of the item is needed. Hidden failures cause operational
surprises which may give rise to accidents. To reduce the risk exposed to hidden
failures, fault-finding (FF) tasks are performed at scheduled intervals to check
the state of items with dormant functions.

The risk of failure


There is a widespread belief that corrective maintenance is always less
economical than preventive maintenance, and all failures can be prevented. As
a result, time-directed maintenance becomes the norm of preventive
maintenance action, motivating the indiscriminate use of overhaul or preventive
replacement procedures in PM programmes. Experience, judgement, vendor
recommendations and “the more the better” syndrome are the common bases Condition-based
for determining the content and frequency of a TD task. This approach to PM maintenance
wastes a lot of resources in doing unnecessary tasks which will not improve
equipment or system availability. Furthermore, PM tasks which involve
intrusion into the equipment (overhaul tasks) are potentially risky. According to
a study on fossil power plants, 56 per cent of the forced outages occurred within
one week after an intrusive type of maintenance task has been performed[2]. 5
The failure rate of an equipment measures the risk of its failure. It is defined
as:
f(t)
h(t) =
R(t)

where f(t) is the failure density, and R(tp ) = ∫ tp f(t)dt is the reliability of an item
at age tp. According to the “bathtub model”, the failure rate, h (·), decreases with
age or usage when the equipment is new – the time during which this applies is
known as the infant mortality period, and it is usually relatively short. This is
followed by a period of constant failure rate (the useful life period), and then a
“wear-out” period. The latter is characterized by a failure rate which is an
increasing function of age or usage. TD tasks are effective in preventing failures
when the equipment is operating within the wear-out region. In such cases, an
overhaul or preventive replacement task will revert the equipment to the “as-
new” condition and, after a brief run-in period, the risk of failure will be
significantly reduced. If the goal of performing TD tasks is to minimize the total
maintenance-related costs, the decision model for determining the optimal
interval between TD tasks is shown below[3]:
C p R ( t p ) + C f [1 − R ( t p )]
C (t p ) =
t p R ( t p ) + M ( t p )[1 − R ( t p )]

where C p is the total cost of a TD task, C f is the total cost of a corrective


maintenance task, M(t p) is the mean time to failure given that the failure
occurred before time tp, and C(tp) is the total cost per unit time if the TD task is
performed when an item is of age tp.
Obviously, performing a TD task will be a waste of resources when:
● h(·) is a non-increasing function of age;
● the cost penalty of a corrective maintenance task, Cf, is not greater than
that of a TD task, Cp.
According to an extensive study conducted in the airline industry, the “bathtub
curve” is not a universal model that applies to all items, as most people have
believed. In fact, as much as 89 per cent of all the airline equipment items do not
have a noticeable wear-out region throughout their service life, and hence these
items will not benefit from TD tasks[4].
JQME Reliability-centred maintenance (RCM)
1,3 Reliability-centred maintenance (RCM) is a structured methodology for
“determining the maintenance requirement of any physical asset in its
operating context”[5]. “The primary objective of RCM is to preserve system
function”[1]. As such, the indiscriminate maintenance tasks which are not cost-
effective in preserving system function should be eliminated[6]. RCM was born
6 in the late 1960s, on the eve of the jumbo jet era. At that time a comprehensive
review of preventive maintenance strategies in the commercial aircraft industry
was performed. The outcome of that review was a new methodology for
structuring PM programmes, and it was defined in MSG-1 (Maintenance
Steering Group-1) for the 747. The MSG-1 methodology was very successful in
preserving the inherent safety and reliability of the 747 at the lowest cost. The
success of MSG-1 led to the development of MSG-2, which was applied in
formulating the PM programme for DC-10 and L1011[4]. The methodology
defined in MSG-1 and MSG-2 was continuously refined and its application has
expanded beyond the civil aviation industry, to cover major military systems (in
the USA), commercial nuclear and fossil power plants, oil platforms, as well as
organizations in other industrial sectors in various parts of the world. In 1975,
the US Department of Defense (DOD) directed that the MSG concept be labelled
“reliability-centred maintenance”.
The RCM methodology is developed on the basis of the following concepts:
● Not all failures are created equally. That is, different modes of failure
may affect the system function differently.
● There are some failures which cannot be prevented by overhaul or
preventive replacement action. The hazard rate of an item with this type
of failure does not change with age.
The RCM methodology uses a systematic technique to rank the criticality of
failure modes and provides guidelines for the selection of applicable PM tasks
that are most cost-effective in preserving system function. A PM task is
applicable when it can accomplish one of the three objectives of performing PM
– prevent failure, detect the onset of failure, or discover hidden failure.
Smith[1] outlined a seven-step process for the analysis involved in a RCM
programme:
(1) system selection and information collection;
(2) system boundary definition;
(3) system description and functional block diagram;
(4) identification of system functions and functional failures;
(5) failure mode and effects analysis (FMEA);
(6) decision-tree analysis;
(7) task selection.
The first phase of the process is concerned with the selection of the system to be Condition-based
covered in the RCM programme. As a general rule, it is not meaningful to start maintenance
the RCM process at the component level because the criticality of a component’s
failures can only be judged on the basis of its impact on delivering the required
system functions. Thus, RCM is usually performed at the system level. The
failure modes to be identified in steps 4 and 5 are those which will cause partial
or complete loss of system functions. The decision-tree analysis in step 6 is 7
shown in Figure 1. It is a systematic procedure for classifying failure modes
into one of three categories:
(1) safety related (including breach of environmental standards);
(2) outage related;
(3) economics related.
Design modification is mandatory to eliminate safety- related failure modes. For
failure modes in categories (2) and (3), the maintenance options include TD
tasks, CBM, FF tasks, run-to-failure (RTF), and design modification.
TD tasks are applicable only if the hazard rate is known to be an increasing
function of age (or usage). When there is insufficient knowledge about the
hazard function and the relevant technology for health monitoring exists, a
CBM task should be considered. It should be noted that, even when a TD task is
applicable, CBM may still be a better option because it eliminates the possibility

Failure mode
(1) Evident
Under normal conditions, do
the operators know that
something has occurred?

Yes (2) Safety No D


Does the failure mode Hidden failure
cause a safety problem?
Return to the decision
tree to ascertain if the
failure is an A, B or C

A Yes No (3) Outage


Does the failure mode result in
Safety problem a full or partial outage of the
plant?

B Yes No C
Minor to insignificant
Outage problem economic problem
Figure 1.
Decision-tree analysis
Source: [1] in a RCM programme
JQME of premature removal of the item from service for PM action. FF tasks are
1,3 applicable for discovering hidden failures which will not create safety hazards
or major economic problems. When it is technically infeasible to perform
effective TD, CBM, or FF tasks, and the failure mode will only have minor
economic penalty, the RTF option will be acceptable.
Unless dictated by mandatory requirements, the RTF option and design
8 modification should always be included as candidates for the final selection of a
PM task. The choice is determined by economic evaluation. Figure 2 shows the
road map for task selection.
The structured analysis process requires the active participation of the
stakeholders, which include the operations and maintenance functions. If
successfully implemented, huge benefits can be obtained from the RCM
programme, including:
● elimination of identifiable safety and environmental hazards;
● control and reduction of O&M costs – this is achieved through more
efficient PM planning and scheduling, more accurate spare parts
identification and stocking, and decrease in corrective maintenance;
● avoidance of outage costs through increased system availability;
● improved operating performance – output, product quality, customer
service;
● greater motivation of those involved in the analysis, and better
teamwork;
● a better understanding of the failure modes and their causes.

Condition monitoring
Apart from the RTF option and design modification, condition-based
maintenance (CBM) is also one of the preferred approaches to PM in a RCM
programme. CBM uses direct monitoring of the mechanical condition, system
efficiency and other indicators to predict the actual time to failure or loss of
efficiency for each item. CBM tasks are performed to serve the following
purposes[7]:
● to determine if a problem exists in an equipment, how serious the
problem is, and how long the equipment can run before failure;
● to detect and identify specific components (e.g. gear sets, bearings) in the
equipment that are degrading, i.e. the failure mode, and to determine the
root cause of the problem – the diagnostic function.
Condition-monitoring techniques can be classified according to the type of
symptoms they are designed to detect. The classifications are as follows[5]:
● dynamic effects, such as vibration and noise levels;
● particles released into the environment;
Is the age reliability relationship Condition-based
(1) for this failure known? maintenance
Yes No

(2) Are there any


applicable TD tasks?

Yes
9

Specify TD tasks No

Are there any applicable CBM


(3) tasks?

Yes

No
Specify CBM tasks

Is there a category D
(4) failure mode?

Yes
Are there any applicable
(5) FF tasks?
No
Yes

Specify FF tasks No

(6) Are any of these tasks effective?

Optional
Yes
No
Is this a category C or
(7) D/C failure mode?

Can a design mode eliminate


(8) the failure mode or its effects?
Yes

No Yes

Specify TD/ Accept run to Design


CBM/FF tasks failure risks modification
Figure 2.
Task selection
Source: [1] guideline
JQME ● chemicals released into the environment;
1,3 ● physical effects, such as cracks, fractures, wear and deformation;
● temperature rise in the equipment;
● electrical effects, such as resistance, conductivity, dielectric strength, etc.

10 The range of condition-monitoring techniques that have been developed for


CBM is surveyed in[5] and[7]. A brief discussion on those which are commonly
used to monitor the condition of mechanical systems follows.

Vibration monitoring
Vibration monitoring techniques can be used to detect fatigue, wear, imbalance,
misalignment, loosened assemblies, turbulence, etc. in systems with rotational
or reciprocating parts, such as bearings, gear boxes, shafts, pumps, motors,
engines and turbines. The operation of such mechanical systems releases
energy in the form of vibration with frequency components which can be traced
to specific parts in the system. The amplitude of each distinct vibration
component will remain constant unless there is a change in the operating
dynamics of the system.
Vibration can be characterized in terms of three parameters, namely,
amplitude, velocity and acceleration. The sensitivity of senors used for
measuring these parameters varies with frequency of the vibration. The
general selection guideline is to use amplitude senors to pick up low frequency
signals, velocity senors in the middle ranges, and accelerometers at higher
frequencies.
In one form of vibration monitoring, readings of the overall vibration energy
between 10 to 10,000Hz are taken from selected points on a machine. These data
are compared to baseline readings taken from a new machine. Alarm limits are
established on the basis of the baseline readings. A fault diagnosis will be
triggered when a reading exceeds its alarm limit. Alternatively, vibration
readings are compared to vibration severity charts to determine the relative
condition of the machine. This approach is known as broadband vibration
trending, and it monitors only the overall machine conditions. The common
microprocessor-based instrumentation for this procedure monitors the root-
mean-square (RMS) level of the vibration.
In the narrowband trending technique, the total energy across a specific
bandwidth of vibration frequencies is tracked to monitor the health condition of
specific machine components or failure modes.
The process of scanning vibration signals across a bandwidth captures
vibration data on the time domain. Such data can be transformed into the
frequency domain so that the vibration at each frequency component can be
measured. The frequency plot that provides a visual representation of each
frequency component generated by a machine is known as the machine’s
vibration signature. When the vibration signatures of a machine at different
times are arranged in chronological order and shown in a cascading manner on
a three-dimensional plot, a waterfall plot of the machine is formed. Anomalies Condition-based
in the machine’s condition can be easily detected by noting that the vibration maintenance
signatures have changed with time.
The guidelines for ensuring that a vibration monitoring programme will not
trigger false alarms and will provide the earliest possible warning of the widest
range of machine faults are given in[8].
11
Process-parameter monitoring
Process parameters cover a wide range of measurements, such as process
efficiency, heat loss, machine temperature, motor current, fluid pressure, etc.
Data on these parameters are normally collected as part of the operational
routine for monitoring system performance. The value of these data can be
further exploited to serve as indicators of the system’s health condition. This
approach to condition monitoring is also applicable to non-mechanical plants
such as heat exchangers, filtration units, pipework and boilers.

Thermography
Thermography uses instrumentation designed to measure emissions of infrared
energy as a means to determine the operating condition of plant machinery.
Anomalies of thermal conditions, such as being hotter or colder than they should
be, are taken as alarm signals of potential problems within the system.
Thermographic techniques are most appropriate to detect problems found in
systems which rely on heat transfer or retention.
Infrared thermometers are designed to measure the surface temperature at a
single point on a machine surface. They can be used to monitor the temperature
of critical parts of plant machinery, such as bearing cap and motor winding, and
to spot check process piping systems. When the infrared emission profile of a
large area needs to be scanned within a short period of time, infrared imaging
will be the applicable technique. The imaging system functions much like a
video camera and the thermal profile of the observed area can be viewed through
the instrument’s optics.
The measurement of infrared emissions is very sensitive to variations of
ambient conditions, such as the amount of airborne particles. Therefore, extra
care must be taken to compensate for the effect of such factors in capturing the
thermal data.

Tribology
Tribology is the field of study relating to the interface between sliding surfaces.
Three tribology techniques are used in condition-based maintenance. They are
lubricating oil analysis, wear particle analysis and ferrography. These
techniques are relatively slow and expensive because the analysis requires the
use of laboratory facilities such as spectrometer and scanning electron
microscope (SEM).
In lubricating oil analysis, samples of lubricating, hydraulic, and dielectric
oils are analysed at regular intervals to determine if they can still meet the
JQME lubricating requirements of their application. When the oil condition reaches
1,3 an unacceptable state, it will be replaced to maintain satisfactory system
operation. Results of the analysis may also form the basis of the decision to
change the type of oil used for performance improvement or variety reduction
purposes. Lubricating oil analysis involves the use of spectrographic
techniques to analyse the elements contained in the oil sample. However, it
12 must be supplemented with other diagnostic procedures in order to identify the
specific failure mode which may have caused the observed degradation of the
oil condition.
Wear particle analysis provides direct information about the wearing
condition of the machine. This information is derived from the study of
particle shape, composition, size and quantity. There are two basic types of
wear particle analysis. In the first type of analysis, the solids content of the
machine lubricant – the quantity, size, and composition of particulate matter in
the lubricating oil – is monitored on a routine basis to detect whether or not it
changes with time. In the second method, particulate matters in each
lubricating oil sample are analysed to identify the type of wear (rubbing wear,
cutting wear, rolling fatigue, combined rolling and sliding wear, or severe
sliding wear) that has left its clues in the sample.
Wear particle analysis using spectrographic techniques is limited to the
study of particulate matters with a size not exceeding 10µm. Ferrographic
techniques have to be used to study larger particles separated from an oil
sample. To measure the concentration of wear particles in an oil sample using
ferrographic techniques, the oil is allowed to flow down an inclined slide,
passing through a strong, magnetic gradient field so that the particles are
separated by order of decreasing size. If non-magnetic particles are to be
separated as well, the oil sample has to pass through a series of calibrated
membrane filters instead of a magnetic field. Using this method, wear
condition can be monitored to detect trends indicating abnormal wear. When
abnormal wear is found, ferrograms, which are microscopic images of the wear
particles lined up by decreasing size, are prepared to visually analyse the wear
particles so that the location and nature of the wear can be identified in time to
prevent catastrophic damage[9].
The limitations of tribology analysis in a condition-based maintenance
programme are: high equipment costs, being a laboratory-based procedure,
reliance on acquisition of accurate oil samples and skills needed for proper
interpretation of data.

Visual inspection
Visual inspection of plant equipment offers a simple method to detect problems
such as leaks, loose mountings, structural cracks, etc. Such inspection can be
performed as the maintenance crew collects the regular data needed by any
other condition-monitoring procedures. Hence, the incremental cost of carrying
out visual inspection is often minimal.
Apart from the above techniques, there are many other non-intrusive Condition-based
methods which have been developed to detect various types of impending maintenance
failures in mechanical systems. These include analysis of acoustic emissions,
eddy currents, ultrasonic waves, etc.

Decision making in condition-based maintenance


There are three types of decisions which need to be made in condition-based 13
maintenance:
(1) selecting the parameters to be monitored;
(2) determining the inspection frequency;
(3) establishing the warning limit (the trigger).
Selecting what to monitor depends on factors such as the type of plant
facilities to be covered, the availability of reliable surveillance technology,
investment in instrumentation and training, manpower requirement and
operating costs. It should be noted that an incipient functional failure may
show several types of symptoms, each of which may become detectable at
different stages of degradation of the unit. For example, the earliest warning
signal of a bearing failure is in the form of changed vibration characteristics.
This is followed in sequence by increased particle content in lubricating oil
samples, emission of audible noise and heat built up on the bearing caps.
Symptoms which show up at later states provide a shorter grace period to the
occurrence of functional failure. All these symptoms and their applicable
condition-monitoring techniques should be considered in choosing the most
cost-effective CBM procedure.
If equipment condition is not monitored continuously the interval between
inspections has to be determined, so that action can be taken in time either to
prevent functional failures or to minimize the consequences of those which
cannot be prevented. If this interval is too short, there will be over inspection.
If it is too long, it will run the risk of having more unplanned outages. An
approach to establishing inspection intervals relates it to the time gap T
between the earliest instant when symptoms of the impending failure becomes
detectable, to the moment when catastrophic failure occurs. The planned
inspection interval plus the expected preparation time for any necessary
preventive action should not exceed one half of T. The uncertainty associated
with the estimation of T that is found in many practical situations complicates
the decision process. Some of the decision models designed to address this type
of inspection problem will be introduced in the following parts of this article.
A condition-based maintenance programme may use both static and
dynamic warning limits as triggers for diagnostic and replacement action.
Static limits are pre-selected thresholds of the measured data. The ISO
vibration severity charts are examples of static limits for vibration monitoring.
While CBM tasks triggered by static limits are easier to administer, they do not
JQME have the diagnostic power for predicting when the alarm will be reached.
1,3 Dynamic limits, on the other hand, are used to monitor the rate of change of the
measured parameter. If a CBM procedure uses dynamic warning limits, the
rate of change of the measured parameters is considered more important than
the actual value. Obviously, the use of dynamic limits will enhance the
automatic diagnostic capabilities of a CBM procedure.
14
Decision models for CBM
A major class of decision problems in CBM relates to the issue of inspection. It
deals with situations in which inspection tasks are performed at periodic
intervals to determine the true state of the equipment. Based on the result of
each inspection, two decisions will be made:
(1) what maintenance action to take – to replace or repair the system to a
specific state, or to leave it as is;
(2) when the next inspection is to take place.
Valdez-Flores and Feldman[10] survey the decision models for the inspection
problem of single unit systems. Some of these models are briefly introduced
below.
Luss[11] proposes a model with these assumptions:
● perfect inspection;
● opportunities for inspection occurs every ∆τ time units;
● the system’s level of deterioration has a discrete state space, and the
holding time in the various states is exponentially distributed;
● infinite planning horizon;
● a failure can only be detected by an inspection;
● occupancy costs are non-decreasing with degree of deterioration;
● repair costs are the same for all states except failure;
● repair at failure is more costly than repair immediately after an
inspection.
A control limit policy with state dependent inspection interval is developed to
minimize expected cost per unit time. The decision variables of the model are:
● the critical state triggering a preventive repair;
● the time to the next inspection τ i where i is the observed state at
inspection.
The necessary condition for existence of an optimal solution as well as an
iterative procedure to solve the model are provided.
An extension of Luss’s model is given by Sengupta[12]. In this extension, the
repair cost is an increasing function of deterioration and delayed repair/
replacement action is allowed. An iterative algorithm to find the solution is also
presented.
Zuckerman[13] develops a model to deal with problems having these Condition-based
characteristics: perfect inspection; fixed inspection intervals; shocks occur in a maintenance
HPP and the degrees of damage caused by successive shocks are IID on ℜ;
infinite planning horizon; a failure is discovered only at inspection; the repair
and operating costs are non-decreasing with degree of deterioration. A control
limit policy for minimizing expected cost per unit time is presented. The
decision variable in this model is the critical state triggering a preventive repair. 15
The necessary condition for existence of an optimal solution is also specified.
Although a procedure to solve the model is not given, it notes that the difficulty
depends on the structure of the survival function of the system and the
distribution of the damage induced by the shock.
Rosenfield[14] considers a slightly different scenario with these conditions:
perfect inspection; the system deteriorates according to a discrete-time Markov
process; the operator has three choices – repair, no action, or inspection; repair
and operating costs are non-decreasing with degree of deterioration. The
objective function is to minimize long-run expected cost per unit time, or
expected total discounted costs. The conditions of system parameters such that
a four-region policy would be optimal are also given.
Three inspection policies are considered by Kander[15]: pure checking,
truncated checking, and checking plus monitoring. In these policies, the next
inspection time is determined on the basis of the most recent state of the system
and discrete deterioration levels are assumed. Under the pure checking policy,
the next checking is always scheduled when the item is found in an unfailed
state. The truncated checking policy specifies that the item will be replaced
when it has deteriorated to one of its late states. The third policy is similar to the
truncated checking policy, except that a continuous monitoring procedure is
initiated when the item is found in a late state. The objective function of these
models is to minimize long-run expected cost per unit time. A procedure for
obtaining the optimal solution is not provided.
Other types of inspection model have also been developed to handle problems
like imperfect inspections[16], hazardous inspections in which each inspection
will cause immediate failure or increase the failure rate[17] and inspection of a
stand-by system[18].

Proportional hazards modelling


In the classical approach, the risk of an item’s failure is determined by reference
to its age or usage only. It is recognized that equipment condition monitoring
procedures, such as vibration monitoring and spectrographic analysis of oil
samples, also provide a wealth of information that can make the assessment of
such risk more reliable. Proportional hazards modelling (PHM), a multivariate
regression analysis procedure first introduced in[19], offers the method for this
assessment. Early applications of the PHM procedure were mainly in the
medical field. Attempts to use the procedure to estimate the reliability of
aircraft, marine and locomotive engines first appeared in the open literature
in[20-22].
JQME The additional information acquired from condition monitoring tasks, and
1,3 analysed by the PHM procedure, can be put to good use in the condition-based
maintenance decision process. Makis and Jardine[23] provide a control theory
approach which incorporates PHM risk assessment in the cost consideration, to
optimize replacement decisions. In that procedure, the level of deterioration
determined by the PHM procedure is discretized and the transition from one
16 level to another is modelled as a semi-Markov process. Inspections are
performed at fixed time intervals. The decision at each inspection is either to
replace immediately, or to plan to continue operating for another x inspection
intervals. The value of x is state and age dependent.

Concluding remarks
CBM is gaining acceptance as a preferred approach to preventive maintenance.
The applicable technologies for capturing data on equipment conditions and the
decision tools for analysing such data are the enabling factors for implementing
CBM. The decision models surveyed in the last two sections of this article are
attempts to address the inspection problems in various scenarios. In many of
these models a system’s level of deterioration is considered to have a discrete
state space and the transition from one state to another is modelled as either a
semi-Markov process or a Poisson process. Decision models featuring a
continuous state space will be welcome additions to the tool kit for inspection
decision making. It is also recognized that the currently available decision
models are commonly designed to optimize a single criterion such as minimizing
the total maintenance related costs, or maximizing equipment availability. Since
decision making in practice is often characterized by the need to satisfy multiple
goals, the formulation of multi-criteria decision models is another worthwhile
topic of future research work in inspection problems.

References
1. Smith, A.M., Reliability-centered Maintenance, McGraw-Hill, New York, NY, 1993.
2. Corio, M.R. and Costantini, L.P., “Frequency and severity of forced outages immediately
following planned or maintenance outages”, Generating Availability Trends Summary
Report, North American Electric Reliability Council, 1989.
3. Jardine, A.K.S., Maintenance, Replacement and Reliability, Pitman/Wiley, London, 1973.
4. Nowland, F.S. and Heap, H.F., Reliability-Centered Maintenance, Department of Defence
Documentation Centre, Defence Logistics Agency, Alexandria, 1978.
5. Moubray, J., Reliability Centred Maintenance, Butterworth-Heinemann, Oxford, 1990.
6. Anderson, R.T. and Neri, L., Rel iabil ity-centered Maintenance: Management and
Engineering Methods, Elsevier Applied Science, London, 1990.
7. Mobley, R.K., An Introduction to Predictive Maintenance, Van Nostrand Reinhold, New
York, NY, 1990.
8. Serridge, M., “What makes vibration condition monitoring reliable?”, Noise & Vibration
Worldwide, September 1991, pp. 17-24.
9. Nunnari, J.J. and Dalley, R.J., “An overview of ferrography and its use in maintenance”,
Tappi Journal, Vol. 74 No. 8, 1991, pp. 85-94.
10. Valdez-Flores, C. and Feldman, R.M., “A survey of preventive maintenance models for Condition-based
stochastically deteriorating single-unit systems”, Naval Research Logistics Quarterly,
Vol. 36 No. 4, 1989, pp. 419-46. maintenance
11. Luss, H., “Maintenance policies when deterioration can be observed by inspection”,
Operations Research, Vol. 24 No. 2, 1976, pp. 359-66.
12. Sengupta, B., “Maintenance policies under imperfect information”, European Journal of
Operations Research, Vol. 5, 1980, pp. 198-204.
13. Zuckerman, D., “Inspection and replacement policies”, Journal of Applied Probability,
17
Vol. 17, 1980, pp. 168-77.
14. Rosenfield, D., “Markovian deterioration with uncertain information – a more general
model”, Naval Research Logistics Quarterly, Vol. 23 No. 3, 1976, pp. 389-405.
15. Kander, Z., “Inspection policies for deteriorating equipment characterized by N quality
levels”, Naval Research Logistics Quarterly, Vol. 25 No. 2, 1978, pp. 243-55.
16. Christer, A.H. and Waller, W.M., “Delay time models of industrial inspection maintenance
problems”, Journal of the Operational Research Society, Vol. 35 No. 5, 1984, pp. 401-6.
17. Chou, C.K. and Butler, D.A., “Assessment of hazardous inspection policies”, Naval Research
Logistics Quarterly, Vol. 30 No. 1, 1983, pp. 171-7.
18. Nakagawa, T., “Optimum inspection policies for a standby unit”, Journal of Operations
Research Society, Vol. 23, 1980, pp. 13-26.
19. Cox, D.R., “Regression models and life tables”, Journal of Royal Statistical Society, B,
Vol. 34, 1972, pp. 187-220.
20. Jardine, A.K.S. and Anderson, M., “Use of concomitant variables for reliability estimation”,
Maintenance Management International, Vol. 5 No. 2, 1985, pp. 135-40.
21. Jardine, A.K.S., Anderson, P.M. and Mann, D.S., “Application of the Weibull proportional
hazards model to aircraft and marine engine failure data”, Quality and Reliability
Engineering International, Vol. 3 No. 2, 1987, pp. 77-82.
22. Jardine, A.K.S., Ralston, P., Reid, N. and Stafford, J., “Proportional hazards analysis of
diesel engine failure data”, Quality and Reliability Engineering International, Vol. 5 No. 3,
1989, pp. 207-16.
23. Makis, V. and Jardine, A.K.S., “Optimal replacement in the proportional hazards model”,
INFOR, Vol. 30 No. 1, 1992, pp. 172-83.

Vous aimerez peut-être aussi