Journal Search Engine
Search Advanced Search Adode Reader(link)
Download PDF Export Citaion korean bibliography PMC previewer
ISSN : 1598-7248 (Print)
ISSN : 2234-6473 (Online)
Industrial Engineering & Management Systems Vol.18 No.3 pp.305-314
DOI : https://doi.org/10.7232/iems.2019.18.3.305

Multi-Criteria Decision-Making-Based Critical Component Identification and Prioritization for Predictive Maintenance

Bongjun Ji, Seunghwan Bang, Hyunseop Park, Hyunbo Cho*, Kiwook Jung
Department of Industrial and Management Engineering, Pohang University of Science and Technology, Pohang, Republic of Korea
Production System Research Lab, LG Electronics, Pyeongtaek, Republic of Korea
Corresponding Author, E-mail: hcho@postech.ac.kr
June 23, 2017 September 9, 2018 July 22, 2019

ABSTRACT


Predictive maintenance is currently taking on new relevance. However, recent developments in predictive maintenance focus on selecting the most appropriate algorithm based on the characteristics of a system and data given the critical components. Identifying the critical component has not been difficult because most predictive maintenance has been applied to well-known critical components. However, as the cost of installation for predictive maintenance lowers, it may be desirable to apply predictive maintenance to machines where critical components have not yet been identified, especially machines in small- and medium-sized enterprises (SMEs). In this paper, an identification method for critical components for which predictive maintenance is appropriate is proposed using multi-criteria decision making for application to multi-component, complex machines. This paper proposes a decision-making process considering three different criteria: severity, occurrence, and detectability. The goal is to identify and prioritize critical components for predictive maintenance. The technique for order performance by similarity to the ideal solution (TOPSIS) can take into account decision makers’ preferences. Sensitivity analysis is investigated and discussed. The proposed decision-making approach allows a manufacturer to develop a customized introduction process for predictive maintenance.



초록


    1. INTRODUCTION

    Maintenance is defined as all actions that are appropriate for retaining a machine in, or restoring it to, a given condition (Dhillon, 2002;Duffuaa et al., 2001). In manufacturing, maintenance is inevitable because, regardless of how well a manufacturing machine is designed, the machine degrades over time because its operation causes stress on each machine component and random events can additionally cause machine degradation (Jardine et al., 2006). Maintenance is not only inevitable but also important in manufacturing because it affects productivity, quality of the product, and safety. As a system or machine becomes more complex, more maintenance activity is needed and maintenance department costs increase, which can represent 15 to 70% of total production costs (Bevilacqua and Braglia, 2000). For these reasons, there have been many attempts to improve maintenance, and thus several maintenance concepts have been developed. A maintenance concept can be classified as when maintenance interventions are performed, which can be corrective, preventive, predictive, etc. (Waeyenbergh and Pintelon, 2002).

    Corrective maintenance is a run-to-failure or reactive maintenance strategy (Ahmad and Kamaruddin, 2012). This strategy thus leads to considerable machine downtime and high maintenance costs (Ahmad and Kamaruddin, 2012). However, because responding to all failures before a machine fails is impossible, corrective maintenance is used as a secondary maintenance concept.

    The second maintenance concept is preventive maintenance (also known as preventative maintenance). In the preventive maintenance concept, maintenance is conducted on a planned, periodic, and specific unit interval (e.g., every x cycles, y hours, etc.) to keep a machine in the stated working condition. It is widely used in the manufacturing industry (Coats et al., 2011;Dhillon, 2002).

    Predictive maintenance (PdM) is the third maintenance concept, which uses modern measurement and signal processing methods to accurately diagnose and predict component/machine conditions and remaining lifetime during operation (Dhillon, 2002). It enables maintenance activities such as repairs and replacements to be performed only ‘when it is needed’ or ‘just before failure’ (Andersen and Rasmussen, 1999). Compared with the first two maintenance strategies, the predictive maintenance strategy has the advantage of reducing both the maintenance cost and time required if it is applied well (Ji et al., 2017). Thus, many studies have been conducted on building a reliable predictive maintenance model to minimize maintenance costs and time required using various algorithms such as logistic regression, support vector machines, artificial neural networks, decision trees, and hidden Markov models. Additionally, applications of predictive maintenance are wide, including rotating machinery, electronic components/systems, fuel systems, turbines, engines, etc. (Ahmad and Kamaruddin, 2012). However, most research focuses only on selecting the most appropriate algorithm based on the characteristics of the system and data given a critical component. It has not been problematic to identify critical components because most predictive maintenance has been applied to well-known critical components. However, as the cost of installation for predictive maintenance is lowered, it may be desirable to apply predictive maintenance to machines where the critical components have not yet been identified, especially machines in small- and medium-sized enterprises (SMEs).

    The remainder of this paper is organized as follows. In Section 2, a method background on the existing predictive maintenance introduction process and on multicriteria decision making is explained. The process of identifying and prioritizing appropriate components for predictive maintenance is proposed in Section 3. Additionally, a case study is presented for a press machine in a SMEs evaluated by the proposed method. In Section 4, our conclusions and future work on component prioritization for predictive maintenance are presented.

    2. METHODOLOGY BACKGROUND

    In this section, we discuss the background methodology required to develop the proposed approach.

    2.1 Predictive Maintenance Introduction

    Even though predictive maintenance is the latest relevant technology, predictive maintenance has not dominated over others. To apply predictive maintenance, continuous monitoring systems should be installed; however, this is expensive because many special devices are required, including a sensor, network. Moreover, data collected is not always clean and without noise (Ahmad and Kamaruddin, 2012). Furthermore, maintenance strategy selection has a multi-objective goal for manufacturers who seek to improve availability/reliability/productivity of machine, quality of the product, and operation safety and to decrease maintenance cost (Faccio et al., 2014); thus, merely selecting predictive maintenance is not necessarily the best approach. Therefore, many researchers have tried to find the most appropriate strategy depending on the situation.

    Azadivar and Shu (1999) presented an approach to select the optimum maintenance strategy for each class of systems considering 16 characteristic factors that play a role in maintenance strategy selection.

    Waeyenbergh and Pintelon (2002, 2004) made many contributions in this research area. They developed a proprietary method to select a maintenance strategy consisting of two steps. The first step is critical component identification, in which they used failure mode and effects analysis (FEMA) to identify critical components, and the second step is selecting an appropriate maintenance strategy using a decision tree that considers technical aspects, economic aspects, and a decision diagram.

    Wang et al. (2007) used a fuzzy analytical hierarchical process (AHP) method to evaluate maintenance strategies considering different factors such as safety, cost, added-value, and feasibility.

    Bashiri et al. (2011) developed a method for selecting the optimum maintenance strategy using a fuzzy interactive linear assignment method. It considers varieties of qualitative and quantitative criteria for maintenance strategy selection, and an expert team was involved in choosing the criteria based on organizational goals and objectives.

    Lee et al. (2014) proposed a maintenance transform map consisting of two axes: system complexity and system uncertainty (Figure 1). According to his transform map, system characteristics should be considered when introducing predictive maintenance. In addition, it is clear that predictive maintenance is a wide-ranging concept that includes e-maintenance, prognosis and health management, condition-based maintenance (CBM), and reliability- centered maintenance (RCM). All four maintenances strategies have commonalities in collecting data and conducting maintenance through prediction using the collected data. They are different only in terms of system complexity and uncertainty. If a system is suitable for applying predictive maintenance, the 5-s approach is adopted as shown in Figure 2.

    Ji et al. (2017) proposed a decision-making process for applying predictive maintenance. He identified three steps and decision-making criteria that should be considered.

    The study of identifying critical components has been addressed in several reliability-centered maintenance papers. In reliability-centered maintenance, critical components are identified based on several criteria. Although a variety of criteria have been proposed to be important, some criteria are commonly used, such as the failure frequency, downtime, maintenance cost, and safety (Wang and Nee, 2009;Hong et al., 1995;Kim and Singh, 1996;Mobley, 2002;Waeyenbergh and Pintelon, 2002;Bridges et al., 2003;Medjaher et al., 2012;Sabouhi et al., 2016;Kandukuri et al., 2016). This is because how often a failure occurs, how long does it take to recover, what cost is needed for recovery, and how much does failure affect safety are common concerns, regardless of the implementation of PHM. Other criteria, on the other hand, are only used in a specific domain. For example, environmental effects are widely used in PHM of power plants (Hong et al., 1995;Bridges et al., 2003;Sabouhi et al., 2016) Given that the failure of a component has serious implications (e.g., radiation could leak from a nuclear power plant), environmental safety is a necessary criterion for evaluating the criticality of a component. In other words, criteria differ depending on the system to which maintenance is applied. In general, productivity, operation, and availability are important criteria in manufacturing context. Table 1

    The proposed criteria are not mutually exclusive. Some criteria can be represented as aggregates of other criteria, as suggested by certain authors. For example, Brahimi et al. (2016) used availability as a criterion, which includes the concept of failure frequency and downtime per occurrence. They defined the critical component as the component that significantly affected availability; however, other researchers have used the failure frequency and downtime as separate criteria. Certain criteria may be described using different terminologies. Failure probability is similar to failure frequency and the mean time to failure (MTTF), whereas downtime is similar to the mean time to repair (MTTR).

    From the literature review of predictive maintenance implementation, it is clear that, even though there are some studies that introduced predictive maintenance methods and identifying critical components in a machine, there is no standardized or widely used way to introduce predictive maintenance. Additionally, if a systematic way is developed, it should be easy for manufacturers to use. There were many quantitative criteria established; however, it is not easy to obtain specific quantitative values in real situations. Hence, the above-mentioned process should be modified and improved. Figure 3

    2.2 TOPSIS

    TOPSIS is a multi-criteria decision-making (MCDM) method (Hwang and Yoon, 1981). That evaluates multiple conflicting criteria when making a decision.

    This MCDM methodology is based on the assumption that the best alternative should be as close as possible to the positive ideal solution and as far as possible from the negative ideal solution (Braglia et al., 2003). TOPSIS has been widely adopted in different areas such as supplier selection in supply chain management (Li et al., 2008), logistics (Behzadian et al., 2012), evaluating performance of competitive aviation firms in business management (Aydogan, 2011), reliability engineering (Almoghathawi et al., 2017), and nuclear safeguard evaluation in safety engineering (Kabak and Ruan, 2010).

    TOPSIS as the method that prioritizes candidate components for predictive maintenance considering multiple criteria.

    Alternatives and the set of criteria are key components in adopting TOPSIS. Alternatives are denoted as A = { A i |   i = 1 , , n } ; in our case, alternative A is the set of the critical component. Ai is the i-th candidate component. The criteria are denoted as C = { C j |   j = 1 , , m } ; in our case, criterion C represents the set of criteria for calculating criticality of components appropriate for predictive maintenance, Cj is the j-th criteria. Let Y = { y i j | i = 1 , , n ; j = 1 , , m } denote the set of criticality ratings of each alternative for each criterion. Additionally, each criterion has its weight, which is denoted as ω = { ω j | j = 1 , , m } . ωj is the weight of j-th criteria. Because the scales are different for each criterion, normalization is needed. Normalization is conducted as follows:

    r i j ( y ) = y i j i = q n y i j 2 ,   i = 1 , , n ; j = 1 , , m
    (1)

    where R = ( r i j ) m × n represents the normalized matrix. The weights are then applied to the normalized matrix.

    v i j ( y ) = ω k r i j ( y ) ,   i = 1 , , n ; j = 1 , , m
    (2)

    Next, the best alternative (positive ideal solution, PIS, Ab) and the worst alternative (negative ideal solution, NIS, Aw) are calculated as below.

    A b = { v 1 + ( y ) , v 2 + ( y ) , , v , + ( y ) } = { ( max 1 i n v i j ( y ) | j J + ) , ( min 1 i n v i j ( y ) | j J ) } A w = { v 1 ( y ) , v 2 ( y ) , , v , ( y ) } = { ( min 1 i n v i j ( y ) | j J + ) , ( max 1 i n v i j ( y ) | j J ) }
    (3)

    The distances between each alternative and the PIS and NIS are found using Euclidean distance. The distance can be calculated as follows:

    D i + = j = 1 m ( v i j ( y ) v j + ( y ) ) 2 , i = 1 , , n D i = j = 1 m ( v i j ( y ) v j ( y ) ) 2 , i = 1 , , n
    (4)

    where the distance between alternative i and the PIS is D i + and the distance between alternative i and the NIS is D i .

    The similarity is calculated through the equation below using D i + and D i . Higher similarity means the smaller distance from the PIS, i.e., farther away from the NIS. Hence, the alternative which has the largest similarity Si is the most preferred. The ranking of alternatives is done in order of similarity.

    S i = D i D i + + D i ,   i = 1 , , n
    (5)

    3. PROPOSED APPROACH

    As mentioned above, in many studies, critical components were identified to select the most appropriate maintenance strategy. In other words, predictive maintenance is not dominant maintenance method. Before the predictive maintenance is applied to a certain component, the determination of whether to implement predictive maintenance or not should be conducted first. In this step, two aspects are considered: economic and technical (Faccio et al., 2014;Waeyenbergh and Pintelon, 2002, 2004). The predictive maintenance should be applied to effective and predictable,

    It is impossible to evaluate all components in a machine as candidate components for predictive maintenance because the number of components in a machine is more than thousands. Hence, candidate component identification may be necessary to minimize the time and cost required for selecting components for predictive maintenance.

    In manufacturing, it is difficult to quantify a cost that reflects the component absence itself (Zha et al., 1998) (Kennedy et al., 2002). Thus, an attempt has been made to measure the criticality of components with the effect of components’ failure. One of the efforts (Ji et al., 2017) borrows the criteria used in Failure Mode and Effect Analysis (FMEA)., a widely used method for measuring the criticality of failure in reliability engineering (Bowles and Peláez, 1995). In this approach, criticality measurement is conducted in terms of severity, occurrence, and detectability. Severity and occurrence relate to the economic aspect and detectability represent the technical aspect. Because severity is defined as “death, injury, occupational illness, damage to or loss of machine or property, damage to the environment, or monetary loss.”, it relates to the economic aspect. Additionally, failure occurs frequently it directly causes loss of monetary. Before the component fails, if the condition or any prior signal from the component indicates the future failure of the component, it can be detected, and technology enables that. Hence, through three concepts, both technical and economic aspects can be covered.

    We propose the TOPSIS approach as a new approach for prioritizing critical components for predictive maintenance from multiple perspectives. It provides the ranking of critical components based on criticality considering multiple criteria. TOPSIS is easy to use compared with other MCDM methods like AHP and Copeland scores because it does not require pairwise comparison.

    3.1 Framework for The Proposed Approach

    To apply the proposed approach, the steps in Figure 4 should be followed. The first step is system failure analysis for selecting both candidate components and criteria for criticality rating. By analyzing system failure using either inductive analysis (FMEA) or deductive analysis (fault trees) candidate critical components for predictive maintenance can be identified. Also in this step, the criteria that evaluate the criticality of a component should be identified. Criteria can be different for different companies because their manufacturing conditions and strategic decisions may not be the same. Therefore, the criteria should take into account relevant and available information, and should also be exclusive to prevent overrating.

    After candidate components and criteria are identi-fied, TOPSIS analysis is conducted from steps 2 to 7. The final output would then be the prioritized components for predictive maintenance.

    3.2 Application of Proposed Approach

    In this section, a real case is introduced to illustrate the proposed approach for applying predictive mainte-nance. The case study was conducted at a Korean company (SMEs) that produces washing machine doors using a semi-H frame single-crank press machine.

    3.2.1 Machine Overview

    The press machine used in this case study is shown in Figure 5, and its specifications are described in Table 2.

    3.2.2 Machine Failure Analysis

    The machine failure analysis was conducted first. The objective of the analysis was to identify candidate components for predictive maintenance and to identify criteria for criticality ratings. For conducting the analysis, machine failure history data were investigated.

    Machine failure data were recorded over more than four years; using these records, we identified failure components. In addition, even though a component may not have failed during the four years of data, it was considered as a candidate component if the component was introduced as a main component in the machine manual. Because main components are managed carefully, failures in these rarely occurred. Finally, 24 components were selected as candidate components for predictive maintenance: the quick die change system (QDC), pump, clutch brake, oil seal, adsorption part, QDC tank, PCB, QDC upper clamp, encoder, safety cut-out, die cushion, balance cylinder, hydraulic hose, clutch oil, press regulator, cushion pin, die lift cylinder, slide gib, main motor, rotary joint, relief valve, key switch, slide motor, motor cooling filter, and slide chain.

    Criteria were selected as shown in Table 3. Three criticality categories and six criteria were selected. Each criterion should be independent unless specific criteria may govern the criticality of a component. Hence, the correlation coefficients between criteria were calculated and show that there was no significant correlation among the criteria. The largest absolute correlation coefficient was smaller than 0.6, indicating that the criteria were sufficiently exclusive

    Severity consist of four criteria: safety, difficulty in recovery, spare part availability, and the possibility of causing secondary damage. Some component failures can cause safety problems for operators. For example, if the die is not held properly or if the safety cut-out does not work properly, it can lead to a serious industrial accident during press operation. Additionally, some components require longer repair/delivery time or higher repair/ replacement cost than other components. Difficulty in recovery reflects these perspectives. Spare part availability also affects the severity of a failure. Not all components can be held because some require extra holding cost and can be burdensome in case of expensive components. Lastly, some component failures may lead to other component failures if not treated in time. To account for this, the possibility of causing secondary damage was measured.

    In the occurrence category, one criterion, frequency, is measured. Other measurements such as the mean time between failures (MTBF) and mean time to failure (MTTF) can be alternatives to frequency; however, a frequency is much easier to collect through the interview, so we selected it as the criterion representing the occurrence category.

    In the case of detectability, prior knowledge is sur-veyed. Prior knowledge of a component indicates how much knowledge the operator has in terms of failure type and the root cause of the failure. Depending on the type and cause of the failure, the data that needs to be collected may vary for predictive maintenance and this is directly related to the success of predictive mainte-nance. In other words, if tacit knowledge is transformed well into explicit knowledge, the probability of success is increased. Hence, prior knowledge can be a good in-dicator for detectability.

    All the criteria should be considered holistically for successful predictive maintenance application.

    3.2.3 TOPSIS Application

    To apply our proposed approach, a domain expert was asked to evaluate the component with selected criteria. Five linguistic scales are used for rating. After all criteria were evaluated, we normalized the rating matrix. Next, we assigned weights to all the criteria; first, we assumed that all the criteria are equally important. In step 5, we determined the best alternative (positive ideal solution, PIS) and the worst alternative (negative ideal solution, NIS). The distances of each alternative from the PIS (D+) and the NIS (D-) were calculated in step 6. Lastly, the ranks of alternatives were determined according to their similarity to the PIS and dissimilarity from the NIS. If similarity to PIS was large, similarity to the NIS was lower the rank was higher. Hence, the final output of TOPSIS was a list of prioritized components as shown in Table 4.

    3.2.4 Sensitivity Analysis

    In Section 3.2.4, we assumed that the weight of each criterion was the same; however, in real cases, the weight may be different from the manufacturing condition. If the manufacturer has criteria that he wants to emphasize, the weights can be modified by the manufacturer. However, most manufacturers struggle in determining the weight of these criteria. Therefore, we conducted a sensitivity analysis by changing the weight and illustrate how this affects the rank of a component.

    We generated ten random weight sets with values between 0 and 1 with a uniform distribution and illustrate the result using a heat map in Figure 6.

    On the horizontal axis, the 24 components are listed and the vertical axis represents their rank values. The components on the horizontal axis were sorted in the order of the smallest sum of ranks. The matrix of the heat map indicates the probability that the component on the horizontal axis is located at the rank of the vertical axis; a darker tone indicates higher the probability of being located at the rank.

    Using the heat map, we can identify the tendency: the smaller the sum of ranks, the higher the probability of being in the upper ranks, even though the weight may change. On the contrary, the greater the sum of the rank, the higher the probability of being in the lower ranks, even though the weight may change. For example, the QDC pump tends to be located at higher ranks even though the weight changed, and the slide chain tends to be located at lower ranks even though the weight changed. According to these results, the QDC pump is highly recommended for application of predictive maintenance; however, the slide chain is not recommended for application of predictive maintenance.

    3.2.5 Evaluation

    The result of the proposed method is compared with human judgment to evaluate the performance. The experts were asked to compare the expected value of applying predictive maintenance to 24 components and prioritize the components by its expected value. In this evaluation, the human experts have sufficient knowledge of the equipment, and have conducted maintenance based on their knowledge and experience. Figure 7 shows the results of the comparison.

    The correlation coefficient of each expert’s ranks versus our result are 0.68 and 0.72. The correlation coefficient of average rank judge by experts versus our result is 0.73, which is higher than above two numbers. This concludes that the proposed approach derives the neural result. The difference between the result of our approach and the judgement of human experts about the importance of main motors and slide motors was relatively large. The reason is that if the certain level of importance in specific criteria is dominated, the component is treated as more important even though the score in others criteria is little bit low. This should be addressed in future work.

    The validation is limited in one case, hence to achieve more statistically valid results, the proposed approach should be applied larger number of equipment.

    4. CONCLUSIONS AND FUTURE WORK

    Predictive maintenance is a maintenance strategy that performs maintenance action based on the monitored condition of a machine. Because it can save maintenance cost and time, it is attracting increased attention as its introduction cost decreases. However, we still cannot apply predictive maintenance to all components in a machine, so we must find appropriate components for applying predictive maintenance and prioritize them. For this reason, we proposed an approach for selecting and prioritizing critical components managed by predictive maintenance. First, we identify the candidate component and its criteria in order to evaluate the suitability and criticality of each component in terms of occurrence, severity, de-tectability. Criteria should be independent, and TOPSIS is applied to prioritize the components. The output of TOPSIS is the ranking of suitable components for predictive maintenance. For calculating a robust ranking, the weights of criteria are changed. The results show that components that have higher ranks tend to be placed at higher ranks; on the other hand, components with lower ranks, tend to be placed at lower ranks. Thus, we can conclude that a component at a high rank should have predictive maintenance applied to it first.

    The proposed an approach can be applied to other machines whose critical components have not yet been identified. Additionally, criteria used in the case study were qualitative rather than quantitative. It is difficult to obtain quantitative data in actual manufacturing environments, especially in SMEs. Therefore, the criteria used in this study can be generally applied in other manufacturing environments as well.

    Using the approach proposed in this paper, there are several interesting future directions of work that can be researched, to investigate systematic approaches for predictive maintenance.

    Firstly, the total list of criteria can be identified for a general and systematic approach. In this case, only available and appropriate criteria are used for evaluating the criticality of a component; in particular, only qualitative criteria are used. However, in cases where well-organized and quantitative data are available for evaluating criticality, quantitative data can be used as criteria. In addition, the proposed criteria may not be appropriate for some manufacturing conditions. It is impossible to cover all manufacturing conditions with the same criteria, so if the total list of criteria is identified, customization in selecting criteria is easy.

    Secondly, further studies can be extended to decide a threshold for applying predictive maintenance. In this paper, only the component rank was calculated for decision support with the assumption that all components are candidates for predictive maintenance. Introducing a threshold will increase the usability of the proposed approach.

    Lastly, other MCDM methods such as the preference ranking organization method for enrichment evaluations (PROMETHEE), the analytic hierarchy process (AHP), and the Copeland score (CS) can be used for result comparison.

    ACKNOWLEDGMENT

    This research was partly supported by the Smart Factory R&D Program of KEIT [10054495 Development of data collection/processing systems capable of adapting manufacturing environments and building a site for demonstrations].

    Figure

    IEMS-18-3-305_F1.gif

    Maintenance transformation map (Lee et al., 2014).

    IEMS-18-3-305_F2.gif

    The 5-s approach (Lee et al., 2014).

    IEMS-18-3-305_F3.gif

    Decision-making process for applying predictive maintenance (Ji et al., 2017).

    IEMS-18-3-305_F4.gif

    Steps for the proposed approach and the outputs of each step.

    IEMS-18-3-305_F5.gif

    A semi-H Frame single-crank press machine.

    IEMS-18-3-305_F6.gif

    Heat map showing the probability of ranks by component.

    IEMS-18-3-305_F7.gif

    Comparison of the result with human experts.

    Table

    Criteria used in measuring criticality

    Specifications of the semi-H frame single-crank press machine

    Selected criteria for evaluating criticality

    TOPSIS output for all components

    REFERENCES

    1. Ahmad, R. and Kamaruddin, S. (2012), An overview of time-based and condition-based maintenance in industrial application, Computers and Industrial Engineering, 63(1), 135-149.
    2. Almoghathawi, Y. , Barker, K. , Rocco, C. M. , and Nicholson, C. D. (2017), A multi-criteria decision analysis approach for importance identification and ranking of network components, Reliability Engineering and System Safety, 158, 142-151.
    3. Andersen, T. M. and Rasmussen, M. (1999), Decision support in a condition based environment, Journal of Quality in Maintenance Engineering, 5(2), 89-102.
    4. Aydogan, E. K. (2011), Performance measurement model for Turkish aviation firms using the rough-AHP and TOPSIS methods under fuzzy environment, Expert Systems with Applications, 38(4), 3992-3998.
    5. Azadivar, F. and Shu, V. (1999), Maintenance policy selection for JIT production systems, International Journal of Production Research, 37(16), 3725-3738.
    6. Bashiri, M. , Badri, H. , and Hejazi, T. H. (2011), Selecting optimum maintenance strategy by fuzzy interactive linear assignment method, Applied Mathematical Modelling, 35(1), 152-164.
    7. Behzadian, M. , Otaghsara, S. K. , Yazdani, M. , and Ignatius, J. (2012), A state-of the-art survey of TOPSIS applications, Expert System with Applications, 39(17), 13051-13069.
    8. Bevilacqua, M. and Braglia, M. (2000), The analytic hierarchy process applied to maintenance strategy selection, Reliability Engineering & System Safety, 70(1), 71-83.
    9. Bowles, J. B. and Peláez, C. E. (1995), Fuzzy logic prioritization of failures in a system failure mode, effects and criticality analysis, Reliability Engineering and System Safety, 50(2), 203-213.
    10. Braglia, M. , Frosolini, M. , and Montanari, R. (2003), Fuzzy TOPSIS approach for failure mode, effects and criticality analysis, Quality and Reliability Engineering International, 19(5), 425-443.
    11. Brahimi, M. , Medjaher, K. , Leouatni, M. , and Zerhouni, N. (2016), Critical components selection for a prognostics and health management system design: an application to an overhead contact system, Proceedings of Annual Conference of the Prognostics and Health Management Society, Denver, USA.
    12. Bridges, M. , Degrave, C. , and Iantchenko, I. (2003), IAEA TECDOC 1383: Guidance for Optimizing Nuclear Power Plant Maintenance Programmes, IAEA, Vienna, Austria
    13. Coats, D. , Hassan, M. A. , Goodman, N. , Blechertas, V. , Shin, Y. J. , and Bayoumi, A. (2011), Design of advanced time-frequency mutual information measures for aerospace diagnostics and prognostics, IEEE Aerospace Conference Proceedings, Big Sky, MT, USA.
    14. Dhillon, B. S. (2002), Engineering Maintenance (1st ed), CRC Press, Florida, USA.
    15. Duffuaa, S. O. , Ben-Daya, M. , Al-Sultan, K. S. , and Andijani, A. A. (2001), A generic conceptual simulation model for maintenance systems, Journal of Quality in Maintenance Engineering, 7(3), 207-219.
    16. Faccio, M. , Persona, A. , Sgarbossa, F. , and Zanin, G. (2014), Industrial maintenance policy development: A quantitative framework, International Journal of Production Economics, 147(Part A), 85-93.
    17. Hong, S. Y. , Jeong, S. I. , and Song, T. H. (1995), Nuclear power plant life time improvement and management program in Korea, IAEA IWG-LMNPP, Austria, Vienno.
    18. Hwang, C. and Yoon, K. (1981), MultipleAttribute Decision Making: Metheods and Applications A Satate-of-the Art Sirvey (Vol. 186), Lecture Notes in Economics and Mathematical Systems, Berlin, Heidelberg: Springer Berlin Heidelberg.
    19. Jardine, A. K. S. , Lin, D. , and Banjevic, D. (2006), A review on machinery diagnostics and prognostics implementing condition-based maintenance, Mechanical Systems and Signal Processing, 20(7), 1483-1510.
    20. Ji, B. , Park, H. , Jung, K. , Bang, S. , Lee, M. , Kim, J. , and Cho, H. (2017), A component selection method for prioritized predictive maintenance, IFIP International Conference on Advances in Production Management Systems, Cham, September, 433-440.
    21. Kabak, Ö. and Ruan, D. (2010), A comparison study of fuzzy MADM methods in nuclear safeguards evaluation, Journal of Global Optimization, 51(2), 209-226.
    22. Kandukuri, S. T. , Klausen, A. , Karimi, H. R. , and Robbersmyr, K. G. (2016), A review of diagnostics and prognostics of low-speed machinery towards wind turbine farm-level health management, Renewable and Sustainable Energy Reviews, 53, 697-708.
    23. Kennedy, W. J. , Patterson, J. W. , and Fredendall, L. D. (2002), An overview of recent literature on spare parts inventories, International Journal of Production Economics, 76(2), 201-215.
    24. Kim, T. and Singh, B. (1996), Guidelines for Implementation of RCM on Safety Systems (No. KAERI/TR--665/96), Korea Atomic Energy Research Institute, Taejon.
    25. Lee, J. , Wu, F. , Zhao, W. , Ghaffari, M. , Liao, L. , and Siegel, D. (2014), Prognostics and health management design for rotary machinery systems - Reviews, methodology and applications, Mechanical Systems and Signal Processing, 42(1-2), 314-334.
    26. Li, W. , Chen, Y. , and Fu, Y. (2008), Combination of TOPSIS and 0-1 programming for supplier selection in supply chain management, In Networking, Proceedings of the 2008 IEEE International Conference on Networking, Sensing and Control, Sanya, China.
    27. Medjaher, K. , Tobon-Mejia, D. A. , and Zerhouni, N. (2012), Remaining useful life estimation of critical components with application to bearings, IEEE Transactions on Reliability, 61(2), 292-302.
    28. Mobley, R. K. (2002), An Introduction to Predictive Maintenance (2nd ed.), Butterworth-Heinemann.
    29. Sabouhi, H. , Fotuhi-Firuzabad, M. , and Dehghanian, P. (2016), Identifying critical components of combined cycle power plants for implementation of reliability-centered maintenance, CSEE Journal of Power and Energy Systems, 2(2), 87-97
    30. Waeyenbergh, G. and Pintelon, L. (2002), A framework for maintenance concept development, International Journal of Production Economics, 77(3), 299-313.
    31. Waeyenbergh, G. and Pintelon, L. (2004), Maintenance concept development: A case study, International Journal of Production Economics, 89(3), 395-405.
    32. Wang, L. and Nee, A. Y. C. (2009), Collaborative design and planning for digital manufacturing, Springer, London.
    33. Wang, L. , Chu, J. , and Wu, J. (2007), Selection of optimum maintenance strategies based on a fuzzy analytic hierarchy process, International Journal of Production Economics, 107(1), 151-163.
    34. Zha, X. F. , Lim, S. Y. E. Y. E. , and Fok, S. C. C. (1998), Integrated intelligent design and assembly planning: a survey, The International Journal of Advanced Manufacturing Technology, 14(9), 664-685.