Software Requirements for Reliability-Centered

Maintenance Application
Iony P. Siqueira
Companhia Hidro Elétrica do São Francisco, CHESF
Ensino Superior e Técnico em Informática, UNIBRATEC

 its implementation is in current use by a task-force (SC-

Abstract—This paper describes some software needs for B3.01) supported by Cigré-Brazil, with contributions from
Reliability-Centered Maintenance (RCM) application to asset many utilities, manufacturers, consultants, research centers
management. Based on the RCM structured approach, a set of and independent system operators, to develop an application
information artifacts and probabilistic data are derived as
necessary for the correct application of the methodology. The
guide of RCM to substation equipments. An ongoing
method and experimental software have been used by a Cigré research contract, financed by ANEEL (The Brazilian
task-force in Brazil, as a supporting tool to develop a guide and Electric Energy Regulating Agency), is supporting the
database for application of RCM to power transformers. A development of information standards for inter-
concurrent research project supported by ANEEL (The organizational asset management, covering RCM data
Brazilian Electric Energy Regulating Agency) and CHESF requirements among users and suppliers of equipments.
(The San Francisco Hidro Electric Company) is also developing
a standard format for exchange of information about RCM
among power companies, equipment suppliers and independent The next part of the paper is a summary of RCM concepts
system operators. Probabilistic optimization of maintenance and steps, related to the analysis of information requirements
frequency is a central requisite of these projects, as this aspect for a software solution. The third part, General
is loosely treated by RCM norms from IEC/ISO and SAE. A Requirements, resumes the informational needs of Failure
stochastic model is suggested and included as a software Mode and Effects Analysis (FMEA), as used by RCM,
requirement.
expressed as requisites in UML, the Unified Modeling
Index Terms—RCM, Reliability-Centered Maintenance, Language standard of the OMG (Object Management
Software Requirements. Group). The fourth part, Activity Requirements, details
software resources to support the decision process of RCM.
I. INTRODUCTION Information Requirements are defined in the fifth part, by
object oriented diagrams describing data structures used by
A mong all contemporary technologies of asset
maintenance, RCM (Reliability-Centered Maintenance)
has expanded its application to practically all industrial
RCM, and data database support needed. The sixth part,
Modeling Requirements, details some stochastic models of
sectors, achieving the status of preferred maintenance failure modes used by RCM as software artifacts. The paper
practice not only in aviation, but on nuclear and electric ends with some Optimization Requirements, in the seventh
industries too. RCM is distinguished by its well structured part, derived from the stochastic models, identifying data
process of analysis and decision, aiming the selection of input and mathematical methods to achieve best maintenance
maintenance activities. The method must be supported by a results.
structured documentation process, for registration,
knowledge management and auditing, in all its steps and All records generated form a documental base sufficient
used data. This requirement has motivated critics to RCM, for certification, knowledge management and auditing from
due to the quantity of forms and related documentation, maintenance engineering. Emphasis is given to support
sometimes seen as bureaucracy. This aspect makes RCM an different statistical (multi criteria) decision models, for
ideal candidate for software support, due to its unified optimization and choice of maintenance frequency.
approach and standardization.
The paper concludes listing several types of reports
It is the aim of this paper to present general requirements needed by a typical user of RCM, such as FMEA, project
for an information system designed to support the control, maintenance plans, etc, in several formats, like
implementation of RCM, according to international SQL/XML/HTML, for intranet and internet publication, as
standards from IEC/ISO and SAE, and with documentation well as standard office formats. Special requirements are
and organizational requisites of ISO 9000 standards. To related to interconnection to traditional CMMS (Computer
adhere to current software technologies, the requirements Maintenance Management Systems) and inter-organizational
should target a multi-user, client-server or multi-layer data exchange for asset management.
solution, to be used stand alone, or concurrently on local and
wide area networks of large corporations. A beta version of II. RELIABILITY-CENTERED MAINTENANCE
Reliability-Centered Maintenance is a structured method
I. P. Siqueira is with Companhia Hidro Elétrica do São Francisco to identify maintenance needs of physical and industrial
(CHESF), Recife, PE, Brazil (e-mail: iony@chesf.gov.br). processes [1,2]. Originated from aeronautical companies, in
1975, and supported by military industry in USA, RCM has  Recovering – from previous data base restoration;
been adopted by nuclear and electric industries, being  Integrity – by backup and compacting data tools.
applied in almost any modern industrial sector, nowadays.
Besides recommending preventive activities, RCM also This list is consistent with modern software requirements
defines a consistent model relating each task to combat for corporate use.
every failure mode. The approach involves answering a
structured set of questions that identify the following treats: IV. ACTIVITY REQUIREMENTS
Any RCM package should work in close relation to the
 Main functions – what the user expects; organization Computer Maintenance Management System
 Functional failures – losses of utility; (CMMS), as shown on the UML Activity Diagram of Fig. 1.
 Failure modes – failure causing events; Note that the task of Maintenance Planning is designed as an
 Failure effects – failure dependent events; RCM macro activity, dependent on the RCM Analysis of the
 Failure consequences – resultant impacts. installation project, and on information supplied by CMMS.
Inside CMMS, this plan generates a Maintenance Schedule
Based on failure consequences (on environment, security, to be followed by execution crews. Failure Analysis occurs
economy or process operation), the method suggests, not only after each failure event, but also after each
through a structured logic, the most applicable and effective maintenance task. Unforeseen failures and statistics should
task to combat each failure mode, among the following trigger execution of RCM analysis, to review, validate or
options: change current maintenance plans. Ideally, these two
packages should function as a unique program. A multi-user
 Time-Directed (TD) – correct before failure; system is recommended for large organizations, with
 Condition-Directed (CD) – detect potential failures; geographically distributed installations, to maximize reuse of
 Failure-Finding (FF) – uncover hidden failures; knowledge and expertise.
 Run-To-Failure (RTF) – repair after failure.
RCM
Following RCM logic, the above order reflects a
decreasing knowledge about the failure mechanism. The last RCM
task is recommended when the previous ones are not cost Analysis

effective, and there is no security or environment issue; Maintenance

otherwise, a project change is mandatory. Planning

Failure Events and Statistics

III. GENERAL REQUIREMENTS Maintenance
Scheduling
Any software to support an RCM process should include
tools to document all its phases. It should supply not only a Maintenance
Execution
guided sequence to RCM logic, but support many
operational (non-functional) requirements such as: Failure Failure
Event Analysis

 Integration – with CMMS and maintenance software;

 Storage – in standard data bases management systems;
 Importation – from standard open formats (XML,etc.);
 Exportation – to standard formats (HTML,XML,RTF); CMMS

 Standardization – follow a recognized RCM standard; Fig. 1 - RCM and CMMS Interaction
 Installation – guided tour for RCM adoption;
 Replication – easy reuse of analysis results; Inside the RCM package, all steps are executed following
 Documentation – unlimited storage of each step data; a structured sequence, as illustrated on the next UML
 Multimedia – inclusion of graphical and sound data; Activity Diagram (Fig. 2). Note that failure modes are
 Help – on-line for each RCM step; initially identified based on physical components, while
 Auditing – native tools for process auditing; failures are initially identified after classification of system
 Security – tools to access control and right permission; functions. This separation, previous to the FMEA step,
 Scalability – from isolated to multi-user operation; allows their concurrent execution, possibly by different
 Management – project follow-up of each item; teams, attaining maximum speed to the process. The FMEA
 Optimization – multi-objective task frequency decision; step reconciles these steps, relating failure modes and
 Planning – block aggregation of maintenance tasks; function failures to their effects and criticality.
 Multiplicity – analysis of several installations/systems;
 Performance – reduced use of network bandwidth;
 Configuration – of data-base and information sources;
 Distribution – of automatic version updates;
 Centralization – to easy network server consolidation;
 Installation
Selection

System
Identification

Component Function
Identification Identification

Failure Mode Functional Failure

Identification Identification

FMEA

Activity
Selection

Frequency
Selection

Fig. 2. RCM UML Activity Diagram

The Activity Selection step, following the FMEA block,

obeys a structured logic, normalized by RCM, as shown on
the UML Activity Diagram of Fig. 8. That logic conducts
the analyst to decide on the most adequate and effective task
to combat each failure mode. This includes traditional Time- Fig. 3. RCM UML Class Diagram
Directed (TD), Condition-Directed (CD) and Failure-
Finding (FF) tasks, as well as proposals for project change. VI. MODELING REQUIREMENTS
Application of RCM is dependent on the development of
V. INFORMATION REQUIREMENTS a suitable model for the dynamic or temporal behavior of
An RCM process uses much data and operating each failure. To provide software support, models can be
information about each project under analysis. Usually the represented by UML state diagrams, from which similar
following items are necessary as input, or to be generated Markov models can be derived.
from the target process:
The behavior of each Failure Mode is usually a stochastic
 Installations – project identification; process, governed by probabilistic events. Modeling of this
 Systems – that compose each Installation; process is usually a requirement for any RCM software. A
 Components – that form each System; general model must represent the dominant concepts of
potential and functional failures. The first, as the detectable
 Failure Modes – of each Component;
event of the start of a functional degradation, also known as
 Functions – performed by each System; a defect; the second as the inability of the item to perform its
 Failures – of each Function; required function. These states can be viewed as slices of
 Causes – for the occurrence of each Failure Mode; areas under the graph (Fig. 4) relating the item resistance to
 Symptoms – generated by each Failure Mode; failure along its operating cycle, delimited by the following
 Actions – possible to combat each Failure Mode; conditions:
 Effects – of each Failure Mode;
 Consequences – impacted by each Effect;  Normal – before a potential failure;
 Activities – selected to combat each Failure Mode;  Defect – between a potential and functional failure;
 Frequencies – of execution of each Activity.  Failure – after a functional failure.

To support the necessary data management, a suitable

static data model should be designed, such as the UML Normal Failure Resistance
Class Diagram of Fig. 3. Observe the close relation among
some classes and the corresponding activities in Fig. 2. This
Deffect
model allows the gathering of many supporting data such as
component failure modes, failure symptoms and causes,
general maintenance actions, as a reference database to be Failure
used by many similar projects. It also helps to structure Time
libraries of previous RCM analysis. Fig. 4. Failure Resistance Curve
 Besides these conditions, the item can be found on a state defects, with no evidence of the exact instant of occurrence.
of preventive or corrective maintenance, according to RCM In consequence, failure rates (f=43), and defect rates
classification. Following this reasoning, five (most probable) (d=14), key parameters of the process, must be inferred
states can be defined for an item, at any time: from other observable variables, using the model. These
variables are visible events and their duration, such as:
1. Normal – apt to play its function;
2. Prevention – under preventive maintenance;  Forced outage frequency (Ff=F43);
3. Repair – unavailable, under repair, after a failure;  Preventive maintenance frequency (Fp);
4. Potential Failure – degradable state, after a defect;  Corrective maintenance frequency (Fc=1/T42);
5. Correction – under rectification after defect detection.  Mean time to maintain (MTTM=T2);
 Mean time to repair (MTTR=T3); and
To derive a (UML) state diagram of this behavior, the  Mean time do correct (MTTC=T5).
following events are identified as causes for state changes:
All of them must be retrieved (by estimation or statistics)
1. Prevent: programmed inspection or maintenance; from the connection of the RCM package to a CMMS
2. Correct: action taken to correct a potential failure; system, as shown of Fig. 1. These are necessary data also for
3. Restore: forced action to correct a functional failure; the optimization of maintenance frequency.
4. Defect: partial or potential functional degradation;
5. Failure: forced functional interruption. VII. OPTIMIZATION REQUIREMENTS
Any maintenance program based on RCM should propose
The term “Correct” refers to a planned event aiming to
a suitable frequency, for each preventive or predictive task.
correct a known potential failure, before its evolution to a
Data and calculations for optimal frequencies should be
functional failure. It differs from a planned “Prevent” event
supported by any RCM software, based on well established
where it is not known if there is a potential failure. The first
theoretical models.
three events are external maintenance events, while the last
two are internal failure events.
Maintenance optimization is achieved by determining
values of task periodicity (T21) and other parameters, such as
These events trigger the transitions between the model
Mean Times to Maintain (MTTM & MTTC) and Repair
states, as shown on the UML State Diagram (Fig. 5), where
(MTTR), which maximize or minimize an objective function.
numbers from 1 to 5 are associated to each state.
Some variables must also obey some restrictions, such as
physical viabilities, available resources and security
Failure (4) - Potential Prevent
Failure
requirements. As MTTM, MTTC and MTTR are limited by
available technology, and assuming they are already at
minimum values, the optimization must be sought by
Defect

adjusting the maintenance frequency. The objective function

Prevent
Restore must reflect the desired result, such as economy, risk, or
(3) - Repair (1) - Normal (2) - Prevention
Prevent
quality of service.

The performance of any stochastic process may be

Correct estimated by a scalar indicator that expresses the cost/benefit
(5) - Correction
of state transition in the system. Each transition ij can be
Fig. 5 - RCM UML State Diagram pondered by a return coefficient (Kij) that measures the
gain/loss for the process for each maintenance event. That is:
A Markov model can easily be derived from this diagram,
allowing the mathematical representation of the stochastic I  K p Fp  K r Fr  K c Fc (1)
process, given their parameters. These can be resumed to
two generalized parameters that define the failure mode where I = scalar indicator or objective function;
behavior of each item: Kp = preventive return rate (per event);
Kc = corrective return rate (per event);
 Defect Rate – probabilistic density of defects or potential Kr = repair return rate (per event);
failure in interval dt, conditioned to absence of defect at Fp = preventive maintenance frequency;
time t; Fc = corrective maintenance frequency;
 Failure Rate – probabilistic density of functional failure Fr = repair frequency.
in interval dt, conditioned to absence of failure at time t
and presence of defect or potential failure at time 0. This is a general expression that can be applied to many
indicators in industry [1]. Among them, the following are
Identification of these parameters in each population of listed as examples, from power system practice:
items is a complex endeavor, in modern industrial systems,
due to the progressive or hidden characteristic of most
 EFO Equipment forced outage; I

 EFD Equipment forced duration; Optimum


Value
LPF Loss of production duration;
 LPP Loss of production probability;
 DNS Demand not supplied; Indicator Prevent
(Kp.Fp)
I
 PNS Production not supplied;
 EOF Equivalent outage frequency;
Fpo
Fp
-C 0 B=1/MTTM
 EOD Equivalent outage duration;
-D
 PDI Production discontinuity index;
 EVC Enterprise variable cost; Correct
(Kc.Fc) Repair


(Kr.Fr)
CVC Client variable cost.
Viable Region
The return coefficients (Kp, Kc and Kr) measure the per-
event cost or benefit contributed to the related indicator.
Suitable statistics can easily be derived from historical data Fig. 6 - Quality Indicators
in a CMMS, to estimate these coefficients.
Note that the positive region of indicator I is formed by
The ideal maintenance frequency that optimizes any of three parcels. The first, Prevent, grows with the increase of
these indicators, taken as an objective function I, can be preventive maintenance frequency, as a cost onus over the
determined by expanding Fc and Fr on expression (1), as desired objective. The second parcel, Correct, decreases
functions of the maintenance frequency Fp. These can be with maintenance frequency, as a benefit brought by
derived from the steady state Kolmogorov equations of a corrective maintenance. The third parcel, Repair, ponder the
Markov model, replicating the structure of the UML state effects of repair tasks on the indicator. This mix is typical of
diagram of Fig. 5. This allows us to build a canonical non- optimization problems, conducting to an equilibrium point
linear programming system such as: among the parcels. The figure also shows how the three
controlled parameters (MTTR, MTTM and Fp) affect the
Minimize the objective function: result. The greater the MTTR, the greater will be the ordinate
of point D, and the frequency that minimizes the indicator.
I  K p Fp  K r Fr  K c Fc The same can be said of MTTM and point B.
(2)
According to classical methods to solve these systems, the
Subject to the restriction: optimum maintenance frequency will be given by a non-
negative real root of the differential equation:
1
 Fp  0 (3)
MTTM . dI
0. (4)
dFp
In these expressions, the repair (Fr) and correction (Fc)
frequencies depend on defect (d) and failure (f) rates,
By substitution, this equation reduces to a quadratic form,
which are functions inherent to process technologies and
with two real roots [1,2,3,4]. Solving it gives the optimum
production environments. Values of MTTM, MTTC and
maintenance frequency, for each failure mode.
MTTR depend on available maintenance technologies, and
also affect these frequencies. The unique controllable
Depending on equations parameters, both roots of equation
parameter is the maintenance frequency. It can be null, if
4 will be negative. In this case, the maintenance frequency
Run-To-Failure is the chosen RCM maintenance task, or
that optimizes the objective function will be zero, at the
greater then zero, in case of Condition-Directed, Failure-
border of the viability region. That is, only a Run-To-Failure
Finding or Time-Directed RCM tasks. The inferior and
strategy is recommended, as any preventive maintenance
superior limits for Fp ((1/MTTM)  Fp ≥ 0) are related to will degrade the objective function.
physical viability, as the maintenance frequency can not be
negative or greater than the inverse of the mean time to do it. If, in addition to availability, there is interest in minimizing
Figure 6 shows a typical plot for these expressions, as a other indicators such as those listed before, a much more
function of maintenance frequency. complex, multi-criteria decision problem (MCDM), will
have to be solved. The same model will still be valid for
each indicator, and a compromise solution will have to be
negotiated among all system results.

As is usual with RCM activities, all results must be

documented, for each failure mode and maintenance task.
The optimum frequency will be a function of the following
data, for each failure mode and task:
 define the software needs.
Maintainability
 Mean time to maintain (MTTM) After its testing in an extensible electrical transmission
 Mean time to repair (MTTR) network, with more then 80 high voltage installations
 Mean time to correct (MTTC) operated by CHESF, it is now being used by Cigré-Brazil
Reliability Task-Force SC-B3.01, to define an RCM Guide for
 Defect rate (d) Substation equipments. A research project to standardize
 Failure rate (f) these requirements among several companies is under way,
Productivity supported by ANEEL and CHESF, and conducted by
 Return rate to maintain (Kp) CESAR, the Advanced Studies and Systems Center of
 Return rate to repair (Kr) Recife, Brazil.
 Return rate to correct (Kc)
Periodicity REFERENCES
 Actual maintenance frequency (Fp) [1] Siqueira, I.P., “Optimum Reliability-Centered
Maintenance Task Frequencies for Power System
Excluding the last item, all other are difficult data to Equipments”, 8th International Conference on
obtain, considering the lack of reliable statistical information Probability Methods Applied to Power System, Iowa,
about operating systems. For new items, good engineering USA, 2004.
estimation and judgment must be used to get initial data. [2] Siqueira, I.P., “Manutenção Centrada na Confiabilidade
Bayesian methods can be used to refine these estimations as – Manual de Implementação”, ISBN 85-7303-566-8,
experience is gained with the process. Accelerated life tests Editora QualityMark, Rio de Janeiro, Brazil, 2005.
may be an option for small and inexpensive components. [3] Siqueira, I.P., « Fiabilité dês Appareillages de
Once used for the first time, successive application of the Protection et son Impact sur lês Performances dês
model will improve the original estimation and results, as Réseaux », Conférence Internationale dês Grands
data from a CMMS system feeds back the RCM analysis Réseaux Electriques, CIGRÉ, Paris, 1996.
step (Fig. 1). [4] Siqueira, I.P., “Processos de Decisão Markovianos em
Sistemas de Proteção em Segurança”, MsC Thesis,
The model can be formatted following the standard forms Univsersidade Federal de Pernambuco, Brazil, 1999.
used by RCM, as shown on Fig. 7, or translated into a simple [5] Siqueira, I.P., “Impact of Protective Apparatus and
spreadsheet or an automated form as input to the RCM Maintenance Scheduling on Power System
software to automate this calculus. Performance”, 4th International Conference on
Probabilistic Methods Applied to Power Systems, Rio
de Janeiro, ELETROBRAS, 1994.
[6] Lyonnet, P., « Optimisation d’une Politique de
Maintenance », Technique & Documentation, Paris,
1993.
[7] Siqueira, I.P., “Análise de Riscos de Adiamento de
Manutenção em Sistemas Elétricos de Potência”, XVI
Seminário Nacional de Geração e Transmissão de
Energia Elétrica, CIGRÉ, SP, 2001.
[8] Almeida, A.T., Souza, F.M.C., Org., “Gestão da
Manutenção – Na Direção da Competitividade”, ISBN
85-7315-273-7, Instituto de Desenvolvimento da
Engenharia de Produção – Recife, PE, 2001
Fig. 7 - RCM Form for Optimization [9] Moubray, J., “Reliability-Centered Maintenance”,
Butterworth-Heinemann, London, 1994.
A research project financed by ANEEL is developing a [10] Norris,J.R., “Markov Chains”, Cambridge University
standard data format, expressed as XML (eXtensible Press,Cambridge,1997.
Markup Language) messages, to encode the information [11] Aven, T., Jensen, U., “Stochastic Models in
exchange among asset management systems of different Reliability”, Springer, New York, 1999.
organizations. These standard can also encode the data [12] Smith, A.M., Hincheliffe, G.R., “RCM – Gateway to
exchange between CMMS and RCM packages, as well as World Class Maintenance”, Elsevier, New York, 2003.
RCM data required from equipment suppliers. [13] Bertling, L., "Reliability Centred Maintenance for
Electric Power Distribution Systems", Doctoral thesis,
VIII. CONCLUSION KTH, Stockholm, Sweden, 2002.
This paper has shown general requirements for Reliability- [14] Hull, E., Jackson, K., Dick, J., "Requirements
Centered Maintenance software application. A statistical Engineering", Springer, London, UK, 2005.
model of equipment defects and failures, and an approach to [15] Booch, G., Rumbaugh, J., Jacobson, I., "The Unified
optimization of maintenance frequency is used as a way to Modeling Language User Guide", Pearson, New York,
USA, 2005.

Fig. 8 - RCM Logic as UML Activity Diagram