Available online at www.sciencedirect.

com

ScienceDirect
Procedia Manufacturing 11 (2017) 1170 – 1177

27th International Conference on Flexible Automation and Intelligent Manufacturing, FAIM2017,

27-30 June 2017, Modena, Italy

Preventive maintenance decisions through maintenance

optimization models: a case study
Sandrina Vilarinhoª, Isabel Lopesª*, José A. Oliveiraª
ªALGORITMI Reasearch Centre, Department of Production and Systems, University of Minho, Guimarães, Portugal

Abstract

Technology has always been a key driver of change in industry, leading enterprises to adopt methods to improve maintenance
decisions and striving for maintenance excellence. This paper reports a procedure to support the planning of preventive
interventions to be integrated in a computerized maintenance management (CMMS) that is discussed considering the difficulties
in its implementation. A basis to get a new CMMS function that allows obtaining the optimal periodicity of preventive
interventions is provided. To this end, failure records based on equipment`s FMEA and reliability study are highlighted to
provide more robust inputs to maintenance models and consequently accurate solutions.
©
© 2017
2017TheTheAuthors.
Authors.Published
Publishedby by
Elsevier B.V.B.V.
Elsevier This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Peer-review under responsibility of the scientific committee of the 27th International Conference on Flexible Automation and
Peer-review under responsibility of the scientific committee of the 27th International Conference on Flexible Automation and
IntelligentManufacturing
Intelligent Manufacturing.

Keywords: Failure Analysis, Maintenance Optimization Models, Preventive Maintenance, Reliability Analysis.

1. Introduction

To achieve excellence in maintenance the balance of maintenance performance, risks, and costs must be taken
into account in order to achieve good quality solutions [1]. This includes developing tactics that maximize the
benefits of maintenance strategies, which are usually classified in two major categories, corrective maintenance
(CM) and preventive maintenance (PM) [2,3]. CM can originate high costs which also include loss of production
incurred due to equipment downtime and, therefore, PM should be performed to reduce these costs whenever it

*
Corresponding author. Tel.: +351 253604762; fax: +351 253604741
E-mail address: [email protected]

2351-9789 © 2017 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
Peer-review under responsibility of the scientific committee of the 27th International Conference on Flexible Automation and Intelligent Manufacturing
doi:10.1016/j.promfg.2017.07.241
 Sandrina Vilarinho et al. / Procedia Manufacturing 11 (2017) 1170 – 1177 1171

reduces the likelihood of the failure occurrence. However, a too high frequency of preventive maintenance
interventions can also result in high costs, once resources may be wasted without been necessary [4,5]. In order to
support PM decision making and replace subjective decisions by objective decisions, maintenance optimization
models have been developed. Maintenance optimization models are applied in order to find a balanced maintenance
solution that is closest to an objective under certain criteria [6].
The work presented in this paper is part of an ongoing project and aims to incorporate a new function in the
existing computerized maintenance management (CMMS) of an automotive company to assist failure analysis and
optimal periodicity definition of preventive interventions, considering costs [5]. This paper reports the
implementation of the procedure to support the planning of preventive interventions, which is discussed considering
the necessary data and its proper organization, and the critical factors for its implementation. The paper begins with
a brief review of time-based maintenance, reliability analysis and maintenance optimization models, in Sections 2.
Then, Section 3 presents the proposed procedure and its application and validation through a numerical example.
Finally, in Section 4 some conclusions are drawn.

2. Literature Review

Time-based maintenance (TBM) also referred in literature as predetermined maintenance is a PM technique [3].
TBM is performed in accordance with established intervals of time or number of units of use, without previous
condition investigation [7]. The general process of TBM implementation includes failure data analysis and
maintenance decision making process [3]. The next subsections address the existing methods to perform failures
analysis and the maintenance models.

2.1. Failure analysis

A failure is defined as an event in which the ability of an item to perform a required function ends [7]. According
to Moubray [8], functional failures identification is followed by the identification of failure modes (which are the
events that causes a functional failure) and the determination of the associated failure effects, that can be understood
as what happens when a failure mode occurs. This is done by performing a failure mode and effect analysis (FMEA)
for each function item. Liu et al. [9] define FMEA as a structured, bottom-up approach that starts with known
potential failure modes at one subsystem level and investigates the effect on the next subsystem level. FMEA may
be the source for virtually all subsequent reliability analyses and assessments because it forces an organization to
systematically evaluate equipment and systems weaknesses, and their interrelationships that can lead to product
unreliability [10].
Item´s reliability is one of the inputs for the application of maintenance optimization models. So, failure analysis
through component reliability study must be performed to this end. It can be carried out using various statistical
tools and, the most common is the Weibull distribution [11]. This probability distribution has been widely used to
model the times to failure of components due to its ability to model various life time distributions, including
increasing, decreasing, or constant hazard rates [3,12]. Data are essential inputs for building decision models that
support evidence-based physical asset management. It must be recognized that mathematical models by themselves
do not guarantee that the right decisions will be made, if the data used do not have the required quality [13]. The
optimization of TBM decisions requires good quality and timely obtained data. Consequently, it is crucial that
failure records, typically maintained in database of CMMSs, are properly treated and organized.

2.2. Maintenance optimization models

Dekker [14] defines a maintenance optimization model as a mathematical model in which the costs and benefits
of maintenance are quantified in order to obtain an optimal balance between them. Maintenance optimization
models aim to evaluate and compare maintenance polices, to determine how often to inspect or to maintain an item
and to help to determine effective and efficient schedules and plans [15]. Dekker [14] concluded that there are
several case studies published which shows that mathematical models are a good way to achieve both effective and
efficient maintenance. However, the author identifies a several factors which may hamper the application of
1172 Sandrina Vilarinho et al. / Procedia Manufacturing 11 (2017) 1170 – 1177

optimization models. The usual problems presented are: data problems (e.g. analyzing data without knowing the
underlying failure mechanisms can lead to wrong results) and the gap between theory and practice (e.g. many
maintenance models have a stochastic nature which is not only difficult to grasp but also difficult to interpret) [14].
Maintenance and replacement problems of deteriorating systems have been extensively investigated in the
literature and, consequently, several maintenance and replacement models have been developed. These models can
fall into some categories of maintenance policies, such as the well-known block replacement policy and age
replacement policy [16]. The first two models concerning these maintenance policies were introduced by Barlow
and Hunter (1960) [12]. From these models, others were created relaxing some assumptions in order to adjust the
models to real systems. Block policy is a replacement practice in which preventive replacements occur at constant
intervals of length tp, regardless of the age of the item, and failure replacements occur as many times as required in
interval (0, tp). Age replacement policy, is a replacement option in which the time when the preventive replacement
occurs depends of the item age. When failures occur, the replacement or repair is made and the time clock is reset to
zero. The next preventive replacement occurs only when the item has been used for a specified age [13]. Ding and
Kamaruddin [6] highlight the gap between academic research and the industrial application, since the industrial
environment is highly complex with different factors and variables that are not completely documented and
analyzed. Hence, a balance point between academic and industrial needs to be identified in order to both sides gain
maximum advantage from the study. It is impossible to completely eliminate the gap, but it can be minimized. The
case study presented in this paper was performed in order to assist manufacturing industry in this sense.

3. Case Study: Implementation process of a maintenance optimization model

Currently, in the automotive company to which the case study refers, the periodicity of preventive maintenance
interventions is defined based on information from equipment suppliers and in the experience acquired by the
maintenance staff. However, these techniques present some fragilities and it is intended that the periodicity of the
interventions be determined more objectively, i.e., using a scientific approach. The main objective is to identify the
optimal balance between preventive and corrective interventions, identifying the optimal preventive interventions
periodicity, considering a given criterion (costs). The procedure proposed to this end considers two crucial
processes: reliability study and application of an optimization method based on a maintenance model [3,13,17]. Fig.
1 shows a conceptual diagram of the proposed procedure.
There is an initial concern to ensure that complete and properly organized failure records and maintenance costs
are maintained, so that it is possible to obtain reliable information to use in both processes. Concerning the
reliability study, the basic purpose is to statistically investigate possible trends in the occurrence of failures and to
obtain the item reliability function, based on the set of collected failure time data. Reliability analysis is generally
performed in five steps presented by Barabadi [17]: collecting data, preliminary analysis of data, model selection,
parameter estimation and model validation. The next process consists on the application of maintenance
optimization models. The maintenance optimization model considered in the procedure is based on age replacement
policy. The model will allow defining the optimal periodicity of the preventive replacement, considering costs as
criterion. All necessary assumptions should be verified before proceeding with the application of maintenance
optimization models.

Failure records Reliability study Reliability Model

Optimization method Optimal preventive
based on a maintenance interventions
+

model periodicity
Maintenance costs

Fig. 1. Processes sequence for the procedure.

The application of the proposed procedure, reported in the following subsections, includes a comprehensive
analysis of the data problems when applying optimization models in order to identify gaps and reflect about
 Sandrina Vilarinho et al. / Procedia Manufacturing 11 (2017) 1170 – 1177 1173

solutions to face them. A rigorous assessment of possible barriers for each process in industrial context and the
identification of methods to exceed them, is crucial for future development of the supporting computer application.
This study also aims to minimize the gap between academic research and practice, once it analyzes data problems in
a more detailed way than most studies about maintenance models available in literature.

3.1. Data record and reliability study

To analyze the applicability of the procedure and identify the needed changes in the company for its application,
a study that considers the replacement of a single-component of a given production equipment was performed. The
selected equipment was a milling machine, and the component name is Index Pin. This subsection presents and
discusses firstly the failure records organization and treatment on the company, highlighting its fragilities and
respective proposed improvements. Next, the reliability analysis is exemplified for the component. From the results
obtained, a maintenance model is applied in order to define its optimum period of preventive replacement.

3.1.1. Failure Records

The company’s CMMS records data related to failure events, however these data are not properly standardized,
complete and organized. This hinders failures analysis and reliability study. One of the fragilities identified is that in
CMMS there is a predefined and very generic list of failure modes. Thus, when the technician fills its report in the
system, he can select one of the failure modes in the predefined list or generate a new one. This practice leads to the
existence of an extensive and redundant failure mode list, allowing the existence of several expressions for the same
failure mode. The non-existence of a complete and organized list of failure modes for each equipment precludes a
properly association between items failure events and its failure modes. Therefore, the identification of the times to
failure associated with a given component type that is inserted in a given machine is just possible through a roughly
form, analyzing information recorded by the technicians in maintenance intervention reports. In order to ensure that
failure records are properly treated and organized, it was decided to adopt a tree structure in the CMMS for failures
record, based on standardized information of respective equipment´s FMEA (Fig. 2). This provides a structured
representation of all the relevant information, in particular, failure modes associated to respective equipment´s
items. The tree structure will include tree levels: group level, subsystem level and component level. For each level
the failure modes will be predetermined and the failure modes of the component level will be connected to the
failure modes of the group level. The record of failure events and preventive maintenance interventions will be
made on this pre-determined structure. Therefore, failures modes will be standardized avoiding the record of the
same type of failure event with different designations such as it is done currently.

Component
Sub-system
System/ Component
Machine
Component
Sub-system
Failure Mode
Component
Failure Mode

Failure Mode Failure Mode

Failure Mode Failure Mode

Fig. 2. Failure tree structure to machines.

The information system should record essential information for the system, subsystem and component. The
causes of the failure at the system level are the failure modes of the subsystem, which in turn their own causes are
the failure modes in the component level [18]. The connection between failure modes of the various levels (system,
subsystem, component) can be done through numeration. At the system level we can have a one-digit numbering
(for example:1; 2), at the subsystem level we can have a two-digit numbering (for example:1.1; 2.1) and at the
1174 Sandrina Vilarinho et al. / Procedia Manufacturing 11 (2017) 1170 – 1177

component level three-digit (for example:1.1.1; 2.1.1). As result, it is possible to visualize the connection between
levels, automatically.
Equipment´s FMEA was started to be performed in the company for few equipment with the aim of equipment
improvement, in a software with no connection with current CMMS. So, this information can be used to define the
failure tree structure to this equipment. A partial tree structure for failures record, based on standardized information
of respective milling machine´s FMEA is depicted in Fig. 3. The component index pin is localized in the subsystem
designed by module 22.

Milling Module 22 Index Pin

Machine

1. Machine stops whith 1.1 Does not detected 1.1.1 Damages pin arm
error message the index pin position

2.1 Does not orientate 2.1.1 Index pin

the PBC correctly mechanism dirty
2. Bad PCB cut

... ...
...

Fig. 3. Part of failure tree structure of the milling machine with emphasis on the component Index Pin.

Each failure event will be recorded associated with a failure mode of a component of the system. Furthermore, this
tree structure will allow the compilation in the CMMS of other relevant failure information to support for instance a
diagnosis support system. To allow the reliability study of a component, the times to failure of a given a socket, i.e.,
an equipment position which at any given time holds a component of a given type [19] should be collected.
Currently in the company, it is just possible to identify times to failure associated with a given component type that
is inserted in a given machine. It is not possible to identify times to failure values associated with a given socket.
Therefore, whenever a machine has more than a component of the same type installed, the database should
distinguish both components. To face this issue, a codification for sockets was proposes. This codification includes
the component type code, the milling machine code and the code defined for the socket, i.e., the component position
in the milling machine. For instance, index pin “8600.860.134–F20–1”, where “8600.860.134”, corresponds to the
component type code, “F20” corresponds to the milling machine code and “1” corresponds to the code of index pin
position in “F20”. The dates of start and end of operation of a given component in a given position must be
recorded. To avoid too many records and interactions with the information system, the start date of operation of a
component can be considered the end date of the previous one. These records will allow creating a history of times
to failure for each socket. As result, item reliability studies will be more reliable. Once there are several milling
machines in the company, this codification will allow taking advantage of a large number of records of times to
failure for different sockets to statistical studies. This proposed change will not be implemented at once, since
equipment’s FMEAs are available only for a few percentage of companies’ equipment. The change will be made
progressively, starting with the most critical equipment in the company. This new records structure will bring
several advantages to the company. The information will be compiled into the CMMS rather than distributed across
different information systems. This, in turn, will allow a more adequate analysis and treatment of information and
facilitate the organization and management of maintenance activities. Thus, this approach intends to allow setting a
complete and organized knowledge base, which will lead to improve maintenance planning, including the
optimization process of preventive maintenance.

3.1.2. Reliability study

Concerning to data collection, as already mentioned, the failure database is not yet organized, to obtain a reliable
set of values of the times to failure of the selected component in a given position of a given milling machine. So, to
exemplify the proposed procedure, a set of values of the times to failure was collected based on information
recorded by the technicians in maintenance intervention reports. In this example, all data corresponds to failures
events. In future, suspended data, including preventive maintenance interventions data, will be considered and the
respective procedure to treat them will be applied. It also must be referred that the records were related to all failure
 Sandrina Vilarinho et al. / Procedia Manufacturing 11 (2017) 1170 – 1177 1175

modes of the considered component. The set of values of the times to failure to exemplify the proposed procedure
corresponds to the most recent five values of failure times data, in days of operation. The successive values of the
times to failure are: 228; 500; 175; 275 and 380. It should be highlighted that, in the presented numerical example,
the set of failure times is smaller than an ideal situation. A larger number of events will be convenient to avoid
compromising the results of the statistical studies.
The preliminary analysis of component failure data consisted on trend analysis of failure data. Trend analysis is
used to test whether the failure process has a trend: decreasing, growing or no trend [13]. A plot of the cumulative
number of failures in function of cumulative operating time (in the socket) was performed and through its analysis,
it is verified that there is no evidence of trend in failures occurrence. To complement this analysis, a trend test, i.e.
the Laplace test, was realized. Specifying a significance level of 5%, on the standard Normal tables, the boundaries
of the critical region are at -1.96 and 1.96. The calculated statistic test is -0.088, which is between -1.96 and 1.96. As
result, the null hypothesis is not rejected meaning that there is no significant trend and failure events are IID
(Independently and Identically Distributed) at 5 % significance level. Therefore, it is appropriate to perform a
Weibull analysis on the data set for the purpose of modeling the failure time distribution.
The selected distribution was the two-parameter Weibull distribution which fits a large number of failure
characteristics of items. To estimate the parameters of the Weibull distribution (η is the scale parameter and β is the
shape parameter), the approach of median rank regression analysis was used. A Weibull probability plot was
obtained and the results of the parameters estimation are η=355.19 and β=2.46. The calculation of the coefficient of
determination yields in R2 =0.9643, and considering that this is an acceptable value, it can be inferred that the two-
parameter Weibull distribution may fit properly the data. The Kolmogorov-Smirnov test was applied in order to
evaluate the goodness fit of two-parameter Weibull distribution. From the Kolmogorov-Smirnov statistic tables, for
a significance level of 5%, the critical value is equal to 0.52. The calculated statistic test is 0.22, which is less than
0.52. As result, the null hypothesis that the two-parameter Weibull distribution fits the data set is not rejected at 5%
significance level. It must be highlighted that, the set of failure times is smaller than an ideal situation and a larger
set of values will allow a best estimates of the Weibull parameters.

3.2. Maintenance optimization model application

For the component under study, the age replacement model to obtain an optimal preventive replacement age was
considered [13]. It is important to note that preventive replacement actions require two necessary conditions. First,
the cost of the corrective replacement must be greater than the preventive replacement cost. Secondly, the
component hazard rate must be increasing [13]. The minimization of the total cost per unit time, C(tp), allows
obtaining the optimal replacement interval.
To determine the optimal preventive replacement age, the cost of a preventive replacement and cost of a failure
replacement of the item have to be considered. From an economic perspective, maintenance operations include both
direct (such for instance, manpower costs, spare-parts costs) and indirect costs (such for instance, non-production
costs, idle operator costs, quality costs) which are more difficult to evaluate. In practice, the cost of a preventive
intervention will be obtained as result of adding costs of: component cost and manpower cost. For the component
cost, the average cost value resulting from several purchases will be considered. Manpower cost in preventive
interventions will be obtained multiplying preventive intervention average duration (in hours) by hourly manpower
cost (in euros per hour). The preventive intervention duration is recorded in work instructions. On the other hand,
the cost of a corrective intervention will be obtained from adding costs of component, manpower cost and non-
production costs. Manpower cost in corrective interventions will result from multiplying average duration of
corrective intervention (in hours) by hourly manpower cost (in euros per hour). The corrective intervention duration
is defined considering the corrective interventions historical, obtaining a mean time to replace the component due to
failure. Non-production costs represent the costs of products that could and have not been produced during
equipment downtime. Then, the estimation of non-production costs will be made multiplying the costs per product
not produced (expressed in euros/product) by the number of products that could and have not been produced (during
equipment downtime). In the company there is not a systematic and complete quantification of maintenance
interventions costs so, it was necessary to infer these values. The maintenance interventions costs above mentioned
are currently being gathered however they are not yet available. Thus, inferred values were assumed in this example.
1176 Sandrina Vilarinho et al. / Procedia Manufacturing 11 (2017) 1170 – 1177

Note that if the value is significantly distant from the real values, the defined periodicity may not be the optimal.
Thus, in this example, it was assumed that the cost of preventive replacement is €100 and the cost of failure
replacement is €500. It also must be noted that, once the estimated shape parameter exceeds 1, i.e., β = 2.46, the
component hazard rate function is an increasing function. This is in accordance with one of the aforementioned
necessary conditions to apply a component replacement model. Therefore, all necessary assumptions are verified to
proceed with the application of the maintenance optimization model.
Using the age replacement model it was obtained a graph of the total cost as a function of different replacement
ages (Fig. 4). Through the analysis of the graphic, it is verified that the minimum cost is approximately between 160
and 190 days, where the total replacement cost per unit age (days) is €0.90. This is the optimal age of component,
i.e., the component should be replaced when it reached the age between 160 and 190 days if it did not fail before.

4
Total costs per unit age

0
0 50 100 150 200 250 300 350 400 450 500 550 600 650 700
Age (Days)

Fig. 4. Total maintenance cost vs. preventive replacement age.

The function can also be analyzed in order to determine an optimum operating range around the optimal time that
does not significantly affect total costs. This will be a relevant contribution to maintenance scheduling since it
allows to assign priority to maintenance actions based on the range. In the presented example, it is clear that a very
acceptable solution is to plan to replace the index pin somewhere between 160 and 190 days. Preventive
replacement age past 240 days is seen to quickly drive up the cost function and, conversely, preventively replacing
earlier than 120 days is seen as over maintenance and increases costs. This presented numerical example allowed
identifying the difficulties in failure analysis and reinforce the idea that data without quality can compromise
statistical studies. It is critical to know the failure modes underlying an item, as well as to have their fault time
recorded properly treated and organized. Otherwise, very poor and insignificant results can be obtained. CMMSs
should be considered not only for storing data but mainly for supporting the processing and analysis of all data. This
certainly will increase the advantages in making decisions based on data and on appropriate models.

4. Conclusions

In this work, a procedure that will be integrated in a CMMS to determine the optimal periodicity of preventive
replacements was presented. The procedure was applied to a specific component in order to analyze and evaluate the
difficulties associated with its use. With this evaluation, the identification of a set of possible problematic situations
that practitioners may face in maintenance optimization models application and corresponding solutions was
identified. The difficulties in the implementation of maintenance models is an issues scarcely discussed in literature.
Solutions were shared to support the automotive company in creating conditions to perform reliability studies and
apply maintenance models. Therefore, this paper contributes to minimize the gap identified in the literature between
theory and practice, analyzing exhaustively the barriers at each stage of the proposed procedure and reflecting on
possible solutions to face them.
The major difficulty was the lack of required records and properly organization for procedure application. The
study highlighted the following limitations: inability to filter time data per component and/or failure mode; failure
 Sandrina Vilarinho et al. / Procedia Manufacturing 11 (2017) 1170 – 1177 1177

information, such as failure modes and causes effects, not standardized; vague and incomplete calculation of the
costs associated with corrective and preventive maintenance interventions, for each equipment´s component. Faced
with these difficulties, a reorganization of actual records and the creation of new ones had to be carried out in the
company CMMS, to ensure rigorous failure time and the availability of maintenance costs. Failure events records
should be supported by a failure tree structure. A new procedure to determine the periodicity of preventive
replacement was settled down to define the periodicity in a more objective manner that it was used in the past by the
company. In the future, it is intended to develop a computer application based on the method described in this paper
and its integration into the current CMMS. A failure diagnosis function will be also designed. This support function
intends to help maintenance technicians to identify failures modes, necessary repair maintenance tasks and spare
parts and, therefore, to improve their efficiency.

Acknowledgements

This research is sponsored by the Portugal Incentive System for Research and Technological Development.
Project in co-promotion nº 002814/2015 (iFACTORY 2015-2018) and has been partially supported by COMPETE:
POCI-01- 0145-FEDER-007043 and FCT – Fundação para a Ciência e Tecnologia within the Project Scope:
UID/CEC/00319/2013.

References

[1] Campbell JD, Jardine AKS, McGlynn J. Asset Management Excellence: Optimizing Equipment Life-Cycle Decisions. second edi. NW: Taylor
and Francis Group, LLC; 2011.
[2] Lie, Chang Hoon; Chun YH. An Algorithm for Preventive Maintenance Policy. IEEE Trans Reliab 1986;35:71–5.
doi:10.1109/TR.1986.4335352.
[3] Ahmad R, Kamaruddin S. An overview of time-based and condition-based maintenance in industrial application. Comput Ind Eng
2012;63:135–49. doi:10.1016/j.cie.2012.02.002.
[4] Tsang A. Condition-based maintenance: tools and decision making. J Qual Maint Eng 1995;1:3–17. doi:10.1108/13552519510096350.
[5] Lopes I, Senra P, Vilarinho S, Sá V, Teixeira C, Lopes J, et al. Requirements Specification of a Computerized Maintenance Management
System – A Case Study. Procedia CIRP 2016;52:268–73. doi:10.1016/j.procir.2016.07.047.
[6] Ding S-H, Kamaruddin S. Maintenance policy optimization---literature review and directions. Int J Adv Manuf Technol 2015;76:1263–83.
doi:10.1007/s00170-014-6341-2.
[7] EN 13306. European Standard: Maintenance Terminology 2010.
[8] Moubray J. Reliability-centred maintenance. 2nd ed. 1997. doi:10.1109/TR.1987.5222285.
[9] Liu H-C, Liu L, Bian Q, Lin Q, Dong N, Xu P. Failure mode and effects analysis using fuzzy evidential reasoning approach and grey theory.
Expert Syst Appl 2011;38:4403–15. doi:10.1016/j.eswa.2010.09.110.
[10] Smith AM, Hinchcliffe GR. Rcm—Gateway To World Class Maintenance. Butterworth-Heinemann; 2004.
[11] Jozwiak IJ. Introduction to the studies of reliability of systems using the Weibull proportional hazards model. Microelectron Reliab
1997;37:915–8. doi:10.1016/S0026-2714(96)00285-5.
[12] Barlow R, Hunter L. Optimum preventive maintenance policies. Oper Res 1960;8:90–100.
[13] Jardine A, Tsang A. Maintenance, replacement and reability: Theory and Applications. Second. London: Taylor & Francis Group; 2013.
doi:10.1017/CBO9781107415324.004.
[14] Dekker R. Applications of maintenance optimization models: a review and analysis. Reliab Eng Syst Saf 1996;51:229–40.
doi:http://dx.doi.org/10.1016/0951-8320(95)00076-3.
[15] Dekker R. On the use of operations research models for maintenance decision making. Microelectron Reliab 1995;35:1321–31.
doi:10.1016/0026-2714(95)99380-2.
[16] Wang H. A survey of maintenance policies of deteriorating systems. Eur J Oper Res 2002;139:469–89. doi:10.1016/S0377-2217(01)00197-
7.
[17] Barabadi A. Reliability model selection and validation using Weibull probability plot - A case study. Electr Power Syst Res 2013;101:96–
101. doi:10.1016/j.epsr.2013.03.010.
[18] Stamatis D. Failure mode and effect analysis: FMEA from theory to execution. ASQ Quality Press; 2003.
[19] Ascher H, Feingold H. Repairable systems reliability: modeling, inference, misconceptions and their causes. New York: 1984.