Professional Documents
Culture Documents
Abstract
To improve the efficiency of reliability-centered maintenance (RCM) analysis, case-based reasoning (CBR), as a kind of artificial
intelligence (AI) technology, was successfully introduced into RCM analysis process, and a framework for intelligent RCM analysis
(IRCMA) was studied. The idea for IRCMA is based on the fact that the historical records of RCM analysis on similar items can be
referenced and used for the current RCM analysis of a new item. Because many common or similar items may exist in the analyzed
equipment, the repeated tasks of RCM analysis can be considerably simplified or avoided by revising the similar cases in conducting
RCM analysis. Based on the previous theory studies, an intelligent RCM analysis system (IRCMAS) prototype was developed. This
research has focused on the description of the definition, basic principles as well as a framework of IRCMA, and discussion of critical
techniques in the IRCMA. Finally, IRCMAS prototype is presented based on a case study.
r 2007 Elsevier Ltd. All rights reserved.
0951-8320/$ - see front matter r 2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ress.2007.03.037
ARTICLE IN PRESS
Z. Cheng et al. / Reliability Engineering and System Safety 93 (2008) 784–792 785
and successfully introduced into RCM analysis process. A analysis, plenty of information needs to be input and much
framework for intelligent RCM analysis (IRCMA) is time is consumed, especially in the stages of identification
studied, in which RCM cases and the basic reliability data of FSIs and FMEA.
of electrical and mechanical items are integrated. The There are generally similar structures in the same
(IRCMA) approach was implemented through an expert category equipment, and some common parts and mechan-
system. ism in the different categories equipment. Namely, there
are similar FSIs in the different equipment. By the
2. RCM analysis process adaptation of RCM analysis information of these FSIs,
the repeated analysis tasks can be reduced and time is
RCM analysis process is as follows [8,10]: saved. RCM analysis efficiency can be improved without
compromising the quality of the analysis [11–15].
(1) Identification of the functionally significant items
(FSI). The FSIs are the items whose failure modes 3. Definition and basic principles of IRCMA
have significant economic, safety or mission effects on
the system. The items may be system, subsystems, By introducing CBR into RCM analysis process, IRCMA
components or parts. The meaning of identification of transforms traditional RCM analysis, which is completely
the FSIs is to identify the functionally significant items conducted by expert for specific equipment, into the retrieval
in system, subsystem, component or part levels. process of the RCM analysis result of similar equipments by
(2) Failure mode and effect analysis (FMEA): The goal of computer, and the adaptation or check process of the RCM
FMEA is to identify functions as well as functional analysis result of the similar equipments by expert. So the
failure modes of the bottom FSI, their failure causes repeated analysis work is avoided or reduced, and the
and effects. efficiency of RCM analysis can be improved.
(3) RCM logic decision. Each failure cause of FSI is
decided by RCM logic decision chart. The specific PM (1) The meaning of IRCMA is to conduct RCM analysis
policy and its interval are ascertained according to on the specific equipments based on the existing cases.
reliability data. (2) The main task of IRCMA is to identify the PM
(4) Combining the PM policy and developing the pre- program for the specific equipment.
ventive program. The optimization of the interval of (3) The goal of IRCMA is to improve RCM analysis
single policy cannot ensure the optimal result of general efficiency and to reduce the costs as well as time of
policies. In order to ensure the efficiency, maneuver- identifying the PM program.
ability and practicability of PM, we need to combine (4) The key to realize IRCMA is computer technology.
the different policies into suitable groups. Generally,
the combination of maintenance policies and develop- The basic principles of IRCMA are as follows:
ment of preventive program should be based on
existing maintenance systems and regular intervals (1) If the structure features, function and application
(date, week, month, season, year), see Fig. 1. environment of equipments are the same or similar,
then their failure modes and measures to prevent failure
From the above RCM process, we may find that are usually close or similar.
complex equipments are generally made up of many FSIs. (2) When the analysts conduct RCM analysis on the
If we conduct RCM analysis on them, we must input FSIs, specific equipments, they often unknowingly refer to
carry out FMEA, conduct RCM logic decision and historical records of RCM analysis on similar items,
combine the PM policy. The characteristics of RCM and achieve the satisfactory outcomes by the adapta-
process are plenty of data, information, and many tables. tion of the similar cases.
Conducting RCM analysis on complex equipments needs (3) There are usually similar structures in the same
many analysts at the same time. In order to conduct RCM equipment category, and some common parts and
Note£
FSI1,FSI2…FSIn denoting the code of FSI
PM program 1A1,1A2…5B3 denoting the code offailure mode
PM1,PM2…PMn denoting the code of PM tasks
mechanism in the different categories equipment. By RCM analysis outcomes will be added to the case base as a
the adaptation of RCM analysis cases of these items, new case. Strictly speaking, the critical realization techni-
the repeated analysis tasks are reduced and RCM ques of the IRCMA include case acquisition and repre-
analysis efficiency can be improved. sentation, formation of query case, case retrieval,
(4) CBR may aid to complete the intelligent retrieval, adaptation and learning. These techniques will be discussed
adaptation and learning of the similar RCM analysis in detail in the following sections.
cases.
5.1. Case acquisition and representation
4. Technology foundations of IRCMA–CBR
The scale of case base and the method of case
CBR is a methodology of problem solving in artificial representation have greater influence on the efficiency of
intelligence (AI), whose fundamental idea is that a solution reasoning [9]. At present we have gathered and captured
is obtained through reference to a catalogue of previous RCM analysis resources and records of about 70 kinds of
problem solving experience. equipment, including artillery, armor, etc. Those data are
The general CBR process includes case retrieval, arranged, classified and integrated to form unique case
adaptation, evaluation and learning [9]. The following base by knowledge engineers and equipment maintenance
chart may describe CBR cycle (see Fig. 2): experts (see Fig. 4).
A whole RCM analysis case ¼ basic features informa-
tion+RCM analysis information.
(1) Description of the problem to be solved.
Some RCM analysis cases and their contents are as
(2) Retrieval of the best matching case(s) from the
follows (see Table 1).
collection of past cases held in the case base.
There are many representation methods for case, such as
(3) Adaptation of the selected case to match the new
first-order logic, predicates, semantic nets, frames, and
problem situation, as the selected case will not be, in
object oriented knowledge representation etc. [7]. Because
general, directly applied to the new situation.
frame representation can combine basic features informa-
(4) Evaluation of a proposed solution fulfilling the
tion with RCM analysis of case, frame representation is
requirements.
applied in this paper.
(5) Learning from both successful and unsuccessful pro-
Frame is made up of frame name and slots, which
posed solutions: if the proposed solution is successful
represent the features of case and include facets [18]. Frame
then it is added to the case base; if it fails then an
representation of equipment RCM analysis case is shown
attempt is made to explain why it failed.
in Table 2.
The framework is made up of function modules to Query case is the collection made up of basic features
realize the IRCMA. information of the equipment to be analyzed. Formation of
The basic input of the framework is the description of query case is to abstract the basic features information of
equipment requiring RCM analysis. The output is the PM the analyzed equipment (system) by user and computer,
requirements of the equipment. IRCMA is driven by a put them into dynamic database and form the equipment
reasoning process based on CBR. (system) query case. According to these features informa-
A framework for IRCMA is shown in Fig. 3, which tion, the cases in case-base matching query case may be
depicts the concept of the IRCMA. retrieved quickly.
From Fig. 3, we can find that according to user’s The basic features information abstraction of the
description of RCM analysis requirement, IRCMA can analyzed equipment is to get the name, application
quickly and accurately conduct the RCM analysis on environment, category and main structures of equipment.
equipment by automatically retrieving the similar cases and
adapting it. The adaptation of similar cases may be 5.3. Case retrieval
completed by identification of FSI, FMEA and RCM
logic decision. If the adapted similar cases are evaluated Case retrieval is the process of finding the case(s)
and proved to be successful in practice, the successful that are closest to the current query case in the
ARTICLE IN PRESS
Z. Cheng et al. / Reliability Engineering and System Safety 93 (2008) 784–792 787
Formation of
equipment query case
equipment RCManalysis
equipment case representation
mechanism
Identification ofFSI
Intelligent
equipment RCManalysis
User
RCM analysiscase
(3) Retrieving the case base according to equipment Setting the weights in the similarity function
category (such as artillery category). By doing this, appropriately can improve the performance of the NN
the similar cases are limited within the same equipment algorithms. Intuitively, more important feature should be
category (artillery category). assigned larger weights than less important features
(4) Retrieving the similar case in the same equipment while totally irrelevant features should be assigned zero
category (artillery category) according to case name. So, weight.
quantities of the similar case name are reduced further. By calculating the feature similarity and weight sums
(5) Nearest neighbor(NN) method. The NN matching of matched features of main structures of the similar
function is a non-parametric classification algorithm cases and the new query case according to similar
based on assumptions of the independence of attributes algorithm, the RCM analysis cases with maximal similarity
in previous cases and the availability of rules and may be found.
procedures for matching. The NN techniques provide a
measure of how similar a previous case is with a given
problem [7]. 5.3.2. Practice case example of case retrieval
Take 122 mm howitzer-type xxx equipment RCM
analysis case as an example, we discuss similarity retrieval
The overall similarity determined by a weighed NN
process of cases based on NN matching function.
matching function is mathematically represented as
Firstly structures and their pattern of equipment are
follows [9]:
presented and query case is formed. Based on the
Pn experience of equipment expert, weight of main structure
wi simðQi ; T i Þ
SIMðQ:T:Þ ¼ i¼1 Pn , (1) name is given 0.3 and weight of their pattern is given 0.7
i¼1 wi
[17]. Structure and their weight of 122 mm howitzer-type
where wi is the weight of feature i, Q is the target case, T is xxx (query case) are shown in Table 3.
the source case, n is the number of feature in each case and
i is an independent feature from 1 to n.
Table 3
Query case of 122 mm howitzer-type xxx
Table 2
Frame representation of equipment RCM case Main structure Pattern
Frame Name: the name of equipment RCM analysis case Barrel 0.3 Monoblock 0.7
Slot1: the description about equipment category construction
facet1: equipment category (artillery category, armor Barrel lock Wedge
category, y, other categories) Counter-recoil Combination
Slot2: the description of used environment of equipment mechanism
facet1: application environment of equipment (island, Elevating gear Tooth sector type
desert y) Deflection mechanism Worm-and-wheel
Slot3: the description about main structures of equipment Wheel buffer unit Torsion bar type
facet1: main structure 1(feature 1, feature value1, weight 1; Wheel brake mechanism Both sides braking
feature 2, feature value2, weight 2y) Braking control device Gas operated
facet2: main structure 2 (feature 1, feature value1, weight 1; Aiming mechanism Independent line of
feature 2, feature value2, weight 2 y) sight type
y Balancing machine Barometric
facet n: main structure n (feature 1, feature value1, weight 1; False cradle Trough
feature 2, feature value2, weight 2 y) Top carriage Toe
Slot4: collection of RCM analysis information Bottom carriage Box type
facet1: FSI structure trees of equipment Carriage side Detached box type
According to 122 mm howitzer-type xxx query case, dent line of sight type/half dependent line of sight
equipment RCM analysis cases retrieved is shown in type) ¼ 0.3+0 ¼ 0.3
Table 4. Match-Value(Feature10) ¼ Match-Value(balancing ma-
Similarity of 122 mm howitzer-type xxx query case chine/balancing machine)+Match-Value(barometric/
and 122 mm howitzer year xxx is firstly calculated. spring) ¼ 0.3+0 ¼ 0.3
According to feature values in table, calculation process Match-Value(Feature11) ¼ Match-Value(false cradle/false
is as follows: cradle)+Match-Value(trough/trough) ¼ 0.3+0.7 ¼ 1
Match-Value(Feature12) ¼ Match-Value(top carriage/
Match-Value(Feature1) ¼ Match-Value(barrel/bar- top carriage)+Match-Value (toe/toe) ¼ 0.3+0.7 ¼ 1
rel)+Match-Value(monoblock construction/mono- Match-Value(Feature13) ¼ Match-Value(bottom car-
block construction) ¼ 0.3+0.7 ¼ 1 riage/bottom carriage)+Match-Value(box-type/de-
Match-Value(Feature2) ¼ Match-Value(barrel lock/ tached box-type) ¼ 0.3+0.7 ¼ 1
barrel lock)+Match-Value (wedge/ Match-Value(Feature14) ¼ Match-Value(carriage side/
whorl) ¼ 0.3+0 ¼ 0.3 carriage side)+Match-Value(detached box-type/detached
Match-Value(Feature3) ¼ Match-Value(counter recoil box-type) ¼ 1.
mechanism/counter recoil mechanism)+Match-Value
(combination/combination) ¼ 0.3+0.7 ¼ 1 According to formula (1), similarity of 122 mm howitzer-
Match-Value(Feature4) ¼ Match-Value(elevating gear/ type xxx query case and 122 mm howitzer year xxx RCM
elevating gear)+Match-Value (tooth sector-type/tooth case is
sector-type) ¼ 0.3+0.7 ¼ 1 P4
Match-Value(Feature5) ¼ Match-Value(deflection me- Match Valueðfeature iÞ
SIM ¼ i¼1 P14 100%
chanism/deflection mechanism)+Match-Value (worm-
i¼1 WeightðiÞ
and-wheel/screw) ¼ 0.3+0 ¼ 0.3
9:8
Match-Value(Feature6) ¼ Match-Value(wheel buffer ¼ 100% ¼ 70%.
14
unit/wheel buffer unit)+Match-Value(torsion bar
type/fold reed) ¼ 0.3+0 ¼ 0.3 In the same way, similarity of 122 mm howitzer-type xxx
Match-Value(Feature7) ¼ Match-Value(wheel brake query case and 152 mm howitzer year xxx RCM case can
mechanism/wheel brake mechanism)+Match-Value be calculated, which is 85%. So, 152 mm cannon howitzer
(both sides braking/both sides braking) ¼ 0.3+0.7 ¼ 1 year xxx RCM case is more similar than 122 mm howitzer
Match-Value(Feature8) ¼ Match-Value(braking con- year xxx, from which RCM analysis outcomes may be
trol device/braking control device)+Match-Value(gas acquired.
operated/hand operated) ¼ 0.3+0 ¼ 0.3 In intelligent RCM analysis system (IRCMAS) proto-
Match-Value(Feature9) ¼ Match-Value (aiming me- type, retrieval process of the similar equipment case based
chanism /aiming mechanism ) +Match-Value(indepen- on NN method is shown in Fig. 7.
Table 4
The retrieved equipment RCM analysis cases
Name pattern 122 mm howitzer-type xxx query case 122 mm howitzer year xxx RCM 152 mm cannon howitzer year xxx
analysis case RCM analysis case
Feature 1 Barrel (name) monoblock Barrel monoblock construction Barrel monoblock construction
Feature 2 Barrel lock wedge Barrel lock whorl Barrel lock wedge
Feature 3 Counter-recoil mechanism combination Counter-recoil mechanism Counter-recoil mechanism
combination combination
Feature 4 Elevating gear tooth sector type Elevating gear tooth sector type Elevating gear tooth sector type
Feature 5 Deflection mechanism worm-and-wheel Deflection mechanism screw Deflection mechanism screw
Feature 6 Wheel buffer unit torsion bar type Wheel buffer unit fold reed Wheel buffer unit torsion bar type
Feature 7 Wheel brake mechanism both sides braking Wheel brake mechanism both sides Wheel brake mechanism single side
braking braking
Feature 8 Braking control device gas operated Braking control device hand operated Braking control device gas operated
Feature 9 Aiming mechanism independent line of sight Aiming mechanism half dependent Aiming mechanism independent line
type line of sight type of sight type
Feature 10 Balancing machine barometric Balancing machine spring Balancing machine barometric
Feature 11 False cradle trough False cradle trough False cradle canister
Feature 12 Top carriage toe Top carriage to e Top carriage toe
Feature 13 Bottom carriage box-type Bottom carriage box-type Bottom carriage box-type
Feature 14 Carriage side detached box-type Carriage side detached box-type Carriage side detached box-type
ARTICLE IN PRESS
790 Z. Cheng et al. / Reliability Engineering and System Safety 93 (2008) 784–792
5.4. Adaptation and learning of case Case learning is a process of adding the new case into
case base, which is also named as case retaining. If the
Case adaptation is a process of transforming a retrieved adapted case is evaluated to be true and feasible, and there
RCM analysis solution into the solution appropriate for are not cases in case base whose similarity measure exceeds
the current problem, which is completed by rule-based the threshold provided in advance, such as 0.9, then the
reasoning and rules in rule base. For those cases that are case will be added into case base. Otherwise, that case is
found to be similar, after case adaptation, information discarded. With the accumulation of new RCM cases, the
from those ‘‘generic’’ cases is applied in the specific case base will become much larger than before and the
situation being analyzed. ability to aid RCM analysis will boost up gradually.
There are two methods of case adaptation, namely
automatic adaptation and manual adaptation. Significa- 6. A case study: a practical application of IRCMAS
tion of automatic adaptation is that the retrieved similar
cases are automatically adapted and replaced by computer Based on the above theories and techniques study, an
according to user question. While signification of manual IRCMAS prototype has been developed and a case is
adaptation is that the retrieved similar cases are adapted studied.
according to the experience of analysts.
If one generic case might be very similar with situation 6.1. Function of the system
being analyzed, but it might be different that calls for
different strategies and intervals, then those maintenance The main functions of IRCMA include project manage-
strategies and intervals need to be recomputed and ment (new creation project, open project, delete project and
readapted by automatic adaptation or manual adaptation. shut project), analysis tools (the FSIs, FMEA and RCM
In this paper, manual adaptation is mainly applied. logic decision) and case reasoning (case retrieval, case
The technique used for case adaptation is rule-based adaptation, case evaluation and case learning), see Fig. 6.
reasoning, which solves the problem by rules in rule base.
Rule consists of a condition part and an action part. If all 6.2. Working approach in a practical application of the
expressions within the condition part are evaluated to be true, system
the action portion will be executed. The general form of the
rules is as follow: IF /ConditionS THEN /ConclusionS. The working approach and steps of applying IRCMAS
The adaptation for similar case will cost many steps of to conducting RCM analysis for 122 mm howitzer-type xxx
reasoning. For example: are as follows:
Initialized The loss of function caused by this failure
(1) Identification of FSIs. Identification of FSI is the first
fact: mode on its own would not become evident
step of RCM analysis. Complex equipments are usually
to the operating crew under normal
made up of many FSIs. When FSIs of the analyzed
circumstance.
equipment are identified, standard RCM procedure
Target Adapting PM type of the similar equipment
needs input FSIs one by one, and plenty of information
condition: case retrieved from scheduled on-condition
needs to be input and much time is consumed. But the
task to scheduled restoration task.
IRCMAS firstly retrieves the structure tree of FSIs of
The adaptation rules and their orders are: Rule1-Rule2- the similar equipment case from equipment case base.
Rule3. The FSIs of the analyzed equipment are identified by
Rule1: IF (the loss of function caused by this failure adaptation of the structure trees of FSI of the similar
mode on its own would not become evident equipment, and the actual savings are achieved, which
to operating crew under normal is shown in Figs. 7 and 8.
circumstance). (2) FMEA. Based on standard RCM procedure, the
THEN (failure consequences is hidden FMEA of the analyzed FSI needs to be completed by
consequences). identification of the functions, failure modes, failure
Rule2: IF (scheduled on-condition task to detect causes and failure effect of the selected FSI. Plenty of
whether the failure is occurring or about to data needs to be input. But based on the IRCMAS, the
occur is technically infeasible and unworthy FMEA of the analyzed FSI can be completed by the
of doing). adaptation of FMEA information of the similar
THEN (PM task selected is not scheduled equipment case, as is shown in Fig. 8. The above steps
on-condition). are repeated until all the FMEA is conducted for each
Rule3: IF (scheduled restoration task to reduce the FSI of the analyzed equipment.
failure rate is technically feasible and worthy (3) RCM logic decision analysis: RCM logic decision
of doing). analysis ascertains the specific PM tasks and their
THEN (PM task selected is scheduled intervals for each failure causes of the FSI. If
restoration task). RCM logic decision is conducted by standard RCM
ARTICLE IN PRESS
Z. Cheng et al. / Reliability Engineering and System Safety 93 (2008) 784–792 791
Open project
RCM project management
Delete project
Shut project
FSI input
Analysis tool FMEA Fig. 8. Adaptation of FMEA information of the similar equipment case.
Equipment RCM
analysts
RCM logic decision
Case retrieval
Case adaptation
Case Reasoning
Case evaluation
Case learning
Table 5
A segment of prevent maintenance program of 122 mm howitzer-type xxx
1 Counter-recoil Operator Before shooting Monitor the quantity of liquid in counter-recoil Operator unit
mechanism monitoring according to standard. Otherwise, Operator unit
shooting add quantity of liquid in time
2 Scheduled 4 years Check the adjuster and valve device for liquid Repair unit
restoration quantity, and restore their technical states
3 Restoration 8 years Replace counter-recoil liquid of type four Repair unit
been conducted the above three steps, is evaluated to be [3] Deshpande VS, Modak JP. Application of RCM to a medium scale
true, the new case will be retained to case base by industry. Reliab Eng Syst Safety 2002;1(77):31–43.
learning mechanism, and be applied to RCM analysis [4] Deshpande VS, Modak JP. Modeling of uncertainties in reliability
centered maintenance: probabilistic approach. Reliab Eng Syst Safety
of the next equipment. At the same tine, a new RCM 2001;2(71):159–64.
analysis outcome is output in the form of PM program. [5] Martorell S, Sanchez A. The use of maintenance indicators to
See Fig. 10 and Table 5. evaluate the effects of maintenance programs on NPP performance
and safety. Reliab Eng Syst Safety 1999;2(65):85–94.
[6] Rausand M. Reliability centered maintenance. Reliab Eng Syst
7. Conclusion Safety 1998;2(60):121–32.
[7] Shi Y, et al. Application of case-based reasoning technique on gun
RCM analysis is a systematic engineering methodology decision. Comput Mech 1999;9(25):55–9.
to identify preventive maintenance (PM) requirement for [8] Abdul-Nour G, et al. A reliability based maintenance policy: a case
equipments in many countries. RCM analysis process is a study. Comput Ind Eng 1998;35(3/4):591–4.
[9] Watson I. Case-based reasoning is a methodology, not a technology.
much repeated task, and dependent on RCM analysis Knowl-Based Syst 1999;12:303–8.
experience. The IRCMA is feasible to improve efficiency [10] Moubray J. Reliability centered maintenance. London: Butterworth-
and quality of RCM analysis. The study outcomes in this Heinnemann; 1997.
paper will further improve the accuracy and validity of [11] Carretero J, Garcia F, Perez JM, Perez M, Cotaina N, Prete P. Study
RCM analysis, and have great significance on RCM of existing reliability centered maintenance (RCM) approaches used
in different industries. Technical report FIM/110.1/DATSI/00,
popularization and application on military equipment Universidad Politécnica de Madrid, Spain, 2000.
and civilian facility. The next tasks of ours are to further [12] NAVAIR, Guidelines for the naval aviation reliability-centered
perfect the IRCMAS. maintenance. Direction of Commander, Naval Air Systems Com-
We believe that the IRCMAS is a powerful tool for the mand, USA, 1996.
development of RCM programs of physical assets, and has [13] Schlkins N. Application of the reliability centered maintenance
structures methods to ships and submarines. Maintenance 1996;11(3).
a potential future in the RCM market. The IRCMAS is [14] Hall R. Optimizing preventive maintenance using RCM. Main-
substituting the traditional computer aided RCM system tenance 1992;7(4).
(CARCMS) within China’s military industry, and is [15] Ben-Daya M. You may need RCM to enhance TPM implementation.
becoming the new generation of RCM analysis tool for J Qlty of Maintenance Eng 2002;6(2).
weapon systems under development. [16] Jia X, et al. The overview on reliability-centered maintenance. J Mech
Eng Coll 2003;2(14):29–32.
[17] Gao P, Wu S, Cheng Z. The criticality model for equipment based on
References monte carlo method: The 53rd annual reliability and maintainability
symposium, 2006.
[1] Carretero J, Perez JM, Garcia-Carballeira F, et al. Applying RCM in [18] Nilsson NJ. Artificial intelligence—a new synthesis. China Machine
large scale systems: a case study with railway networks. Reliab Eng Press; 2000.
Syst Safety 2003;3(82):257–73. [19] Russell S, Norvig P. Artificial intelligence: a modern approach. San
[2] Deshpande VS, Modak JP. Application of RCM for safety Francisco, USA: Morgan Kaufmann; 1995.
considerations in a steel plant. Reliab Eng Syst Safety [20] Cheng Z. Intelligent RCM analysis. China: PhD’s degree thesis of
2002;3(78):325–34. Mechanical Engineering College, 2006.