Sie sind auf Seite 1von 5

A Realistic Maintenance Model Based on a New State Diagram

Saranga K. Abeygunawardane
Department of Electrical and Computer Engineering National University of Singapore Singapore g0800437@nus.edu.sg
Abstract Equipments that deteriorate over its useful life are subjected to regular inspection and maintenance to prevent failures. These scheduled maintenance activities are frequently modeled using a state diagram due to its simplicity. However, maintenance models based on classical state diagrams provide invalid results when inspection rates are non periodic. This paper discusses the mismatches of the model properties of classical state diagrams with the maintenance situation in the real world. We propose a new state diagram which represents the maintenance situation in the real world. The classical state diagram and the proposed state diagram are analyzed using Markov methods in a numerical example to compute performance measures. These results are compared with the results obtained by conducting Monte Carlo simulation according to practical and realistic inspection and maintenance procedures. It is shown that the classical state diagram can be replaced with the proposed state diagram to obtain an accurate maintenance model, which represents the real situation and provides valid results. Keywords- maintenance model; state diagrams; realistic; Markov methods

Panida Jirutitijaroen
Department of Electrical and Computer Engineering National University of Singapore Singapore elejp@nus.edu.sg widely used to model scheduled maintenance of a system by combining deterioration, inspection and maintenance processes. At first, inspection rates in these state diagrams have been considered as periodic [1-3]. Later on, non-periodic inspection rates have been used in generalized maintenance models to represent the common practice in the real world maintenance [4-7]. In maintenance scheduling and optimization, these state diagrams are converted into Markov models and mathematically solved. Markov models based on state diagrams are very useful to get valid results when inspection rates are constant throughout the deterioration process and this can be easily proved [8]. In the real world, maintenance frequencies are not constant and the common practice is to increase the rate when the system is more deteriorated. However, maintenance models based on classical state diagrams provide inaccurate results when inspection rates are non-periodic [8]. A good attempt can be found in [8] to identify unrealistic properties of maintenance models based on classical state diagrams, when inspection rates are non-periodic and solved using standard Markov methods. Furthermore, a complicated graphical model has been proposed to represent the real world maintenance situation. However, the main drawback of this graphical model is the difficulty of finding analytical solutions [8]. In this paper, a new model based on a state diagram is proposed to represent the real world maintenance situation and to overcome the limitations in classical maintenance models based on classical state diagrams. Unlike the model proposed in [8], this model can be easily solved using standard Markov methods to obtain accurate results. The main focus of this paper is on maintenance models for scheduled maintenance with non-periodic inspection rates. The remainder of this paper is structured as follows. In section II, firstly, the advantages and disadvantages of using state diagrams in maintenance modeling are discussed. Then a classical state diagram, which is the basis of classical maintenance models with non-periodic inspection rates, is described. In this section, it is further discussed that to what extent the classical maintenance model represents the maintenance situation in the real world. Next, a new state diagram is proposed to represent the practical maintenance situation and , the methods of calculating reliability measures of maintenance models are presented in section II. In section

I.

INTRODUCTION

Any equipment or a system with a considerable useful life deteriorates and fails mainly due to usage and aging. Maintenance is the most widely used tool to ensure satisfactorily reliable operation of the equipment or the system. There are three types of maintenance; corrective maintenance, scheduled maintenance or preventive maintenance and predictive maintenance or condition based maintenance. Corrective maintenance repairs the equipment or the system after a failure has occurred. Scheduled maintenance conducts regular inspections and maintenance activities in a predetermined manner to prevent failures. Predictive maintenance is to repair the system when an abnormal condition is detected. In scheduled maintenance, by varying inspection frequencies of the system, the maintenance cost can be minimized while maximizing system reliability. This maintenance scheduling and optimization requires maintenance models to quantify benefits and relate them with costs, in order to assess the impacts of maintenance on system reliability and find optimum inspection intervals. State diagrams have been
This work is supported by Singapore Ministry of Education-Academic Research Fund, Grant No. WBS R-263-000-487-112.

978-1-4244-5721-2/10/$26.00 2010 IEEE

299

PMAPS 2010

III, a numerical example is given to verify that the proposed state diagram can provide a maintenance model, which is in accordance with the actual maintenance situation. A technique to reduce the complexity of the propose maintenance model is also presented in this section. Finally, the conclusions are given in section IV. II. A MAINTENANCE MODEL

1
2 3

State diagrams are widely used to model scheduled maintenance of a system by combining deterioration, inspection and maintenance processes. The main advantage of using state diagrams is its ability to provide more simple and straightforward graphical models for maintenance strategies. Moreover, it can be easily used to build up a mathematical model called a Markov model. These Markov models can be easily solved by standard Markov methods using analytical equations. Although there are several advantages, using classical state diagrams for maintenance modeling can provide inaccurate results under some scenarios. Maintenance models based on classical state diagrams provide accurate results when inspection rates are periodic [8]. However, the results of these classical models are inaccurate when inspection rates are nonperiodic [8]. Therefore, there is a need to propose a better model which represents the real world situation and provide accurate results in order to use state diagrams in future maintenance modeling. A. A Classical State Diagram for Maintenance Modeling Although many maintenance models use a constant inspection rate through out the deterioration process, it is more realistic to use state dependent inspection rates, because when the equipment or the system deteriorates more, inspection frequency should be higher to take necessary maintenance actions before a failure occurs. In this section a classical state diagram with non-periodic inspection rates is described. Fig. 1 shows a classical state diagram with non-periodic inspection rates, which is used for maintenance modeling [8]. States S1, S2, S3 represent the deterioration process, which ends at the failure state F, if no maintenance is carried out. In this model, 1 and 2 are deterioration rates and 3 is the failure rate. When the equipment or the system reaches the failure state F, it is replaced to the as good as new state S1. To increase the lifetime and reliability, inspections and maintenance are carried out throughout the lifetime of the equipment or the system. If the system is in S1, inspections are carried out with a rate of 1. At the inspection state I1 it can be found that the system is still in as good as new condition and therefore no maintenance is required. When the system is at S2 (or S3) inspection rate is 2 (or 3). Inspections at I2 and I3 would result in carrying out maintenance at M2 and M3 respectively, which will improve the system condition only by one stage.

Fig. 1. Example of a classical state diagram [8]

According to the classification of maintenance models, the maintenance models based on classical state diagrams belongs to the category of inspection models [9]. The definition of inspection models given in [9] is Inspection models usually assume that the state of the system is completely unknown unless an inspection is performed. Every inspection is normally assumed to be perfect in the sense that it reveals the true state of the system without error. In the absence of repair or replacement actions, the system evolves as a nondecreasing stochastic process. In general, at every decision epoch there are two decisions that have to be made. One decision is to determine what maintenance action to take, whether the system should be replaced or repaired to a certain state or whether the system should be left as it is. The other decision is to determine when the next inspection epoch is to occur. The assumptions in this definition reasonably agree with the inspection and maintenance situation in the real world. The classical state diagram described previously doesnt perfectly match with this definition due to some poor modeling properties. There are two main mismatches. The first mismatch is regarding the knowledge about the current deterioration state of the system. In the classical model, the deterioration state is always assumed as known even prior to an inspection. Whereas according to the definition of inspection models the state of the system is completely unknown to the operator before an inspection is carried out, unless otherwise the inspection is followed by a previous inspection or a maintenance action. This perfectly agrees with the inspection and maintenance situation in the real world. For example, in the classical state diagram, when the deterioration state is S2 the inspection rate is set to 2 prior to inspections at I2. This inspection rate is not set to 2 in the real world, unless otherwise the operator knows that the system is at S2 either by conducting an inspection at I2 or after a maintenance activity, which improves the system state to S2. Therefore, if S2 is followed by S1, inspection rate cannot be changed and should be remain at 1. However, if S2 is followed by M3, inspection rate should be changed to 2 and the classical state diagram correctly represents this situation. The second mismatch occurs in decision making after inspections. According to the definition of inspection models, after an inspection, it is decided that whether the system is repaired or replaced to a certain state or remains as it is and based on this decision, next inspection rate is determined. This agrees with the common practice in maintenance. However in the classical model, after inspections at I2 and I3, the system is always repaired and the system condition is improved by one deterioration state. At the moment, we let this second mismatch

300

to be relaxed, and assume that inspections at I2 and I3 would always lead to a decision to carry out maintenance. B. Proposed State Diagram for a Realistic Maintenance Model In this section, a state diagram is proposed to model the real world maintenance situation and it is shown in Fig. 2. In this state diagram, the deterioration states S'2 and S" 2 represents the deterioration state S2 in the classical state ' diagram shown in Fig. 1. Similarly, S3 and S" represents S3 in 3 the classical state diagram. All remaining states in the classical state diagram are kept unchanged. Prior to an inspection, that ' is, when the system is at deterioration state S1, S'2 or S3 , the operator does not know that the system is deteriorated. Therefore the inspection rate is kept at 1, because the operator thinks that the system is still at S1.

C. Reliability Measures of maintenance models The two main methods to calculate performance measures of a maintenance model are based on Markov equations and Monte Carlo simulation. If the maintenance model is based on a state diagram, it can be easily solved using Markov methods. If the model is non-Markovian, then it can be solved using Monte Carlo simulation.

After the maintenance actions at M2 followed by inspections at I2, operator knows that system condition is improved to S1 and he inspects the system at a rate of 1. Similarly, maintenance actions at M3 lead to an improvement in system condition and the new deterioration state is S" . When 2 the system is at S" the operator knows that the system is in the 2 second deterioration state and therefore s/he inspects the system at a rate of 2. When the system is at the deterioration state S" followed by S" , operator thinks that the system is still 3 2 at S" and carry out inspections at the same rate of 2. The 2 operator doesnt know that the system is at S" unless after an 3
' inspection at I3. Failure state F is followed by either S3 or S" . 3

When a maintenance model based on a state diagram is converted into a Markov process, it is assumed that the next transition only depends on the current state and the transition rate is constant. Therefore, all transition times are exponentially distributed and all transitions are independent of past behaviors of the system. Due to this assumption the model is easy to solve and there are standard analytical methods to find state probabilities, visit frequencies, mean time to failure, etc. If transition times are not exponentially distributed, semi Markov models can be used [10-12]. In addition, there are some other methods such as device of stages method to convert non-Markovian models with nonexponential distributions into Markov models with constant transition rates [12-14]. However, in this paper it is assumed that all transition times are exponentially distributed and therefore transition rates are constant. Monte Carlo simulation methods are used, when maintenance situation is too complicated to model using state diagrams. According to [8], there are two concepts of conducting Monte Carlo simulation. One concept is redrawing next deterioration time and the next inspection time after each state transition from Si to Si+1 or Ii to Si and this concept is termed as redrawing concept and the simulation results are as same as the results obtained by Markov models based on classical state diagrams [8]. The other concept is drawing next deterioration time only after a change in the deterioration state Si due to deterioration or maintenance and drawing next inspection time based on the decisions after an inspection or maintenance. This second concept, which represents the real world maintenance situation, is termed as non-redrawing concept in [8]. III. NUMERICAL EXAMPLE AND ANALYSIS

In the classical state diagram, there is no transitions from I2 to S2 or from I3 to S3, assuming all inspections at deterioration states 2 and 3 results in maintenance. The same assumption is used in the proposed state diagram, aiming the comparison between the results obtained by the classical state diagram and the proposed state diagram. Therefore there is no need to consider 3 in the proposed state diagram since there is no transition from I3 or M3 to S3.

1 1

1
2

In this section a numerical example is used to check the accuracy of the classical model and the proposed model. The transition rates given in Table 1 are as same as the transition rates used in [8]. Therefore the results obtained in [8] by conducting Monte Carlo simulation for a graphical model which represents the real world maintenance situation can be used to verify the accuracy of using Markov methods for the two models based on the classical state diagram and the proposed state diagram. The models based on the classical state diagram and the proposed state diagram are realized using standard Markov methods to find state probabilities, visit frequencies and mean durations. Some reliability indices such as mean time between failures (MTBF) and mean time to first failure (MTTFF) are also computed. The results for the two models are tabulated in corresponding columns of Table 2, 3, 4 and 5. Then these results are compared with the results obtained in [8] by

1
2
2

1 3

Fig. 2. The proposed state diagram

301

conducting Monte Carlo simulation using non-redrawing concept, which are given in the last columns of Table 2, 3, 4 and 5.
TABLE I TRANSITION RATES [8] Rate 1 2 3 1 2 3 Value[1/year] 0.33 0.29 0.5 0.5 1 1 Rate 1 2 3 1 2 3 TABLE II STATE PROBABILITIES State S1 S2 S3 I1 M2 I2 M3 I3 F Classical Model 0.7326 0.2204 0.0426 0.0010 0.0006 0.0006 0.0002 0.0001 0.0018 Proposed Model 0.6197 0.2931 0.0817 0.0009 0.0005 0.0005 0.0002 0.0001 0.0034 Monte Carlo Simulation [8] 0.61711 0.29641 0.08097 0.00086 0.00046 0.00046 0.00024 0.00012 0.00336 Value[1/year] 360 360 360 12 360 180 MTBF MTTFF from S1 MTTFF from S2

RELIABILITY INDICES [YEARS] Classical Model 46.9283 46.8450 43.8105 Proposed Model 24.4841 24.4008 21.3662 Monte Carlo Simulation [8] 24.7 24.5 23.5

Finally, As can be seen from Table 2, 3, 4 and 5, the results obtained by solving the maintenance model based on the classical state diagram in Fig.1 significantly differ from the results obtained by conducting Monte Carlo simulation. Whereas, the results obtained by solving the maintenance model based on the proposed state diagram in Fig.2 are very much closer to the Monte Carlo simulation results. Although, the classical state diagram provides inaccurate results, the results obtained by the proposed state diagram are as accurate as the results obtained using Monte Carlo simulation. Therefore it is clear that the maintenance model based on classical state diagram does not represent the real world maintenance situation. On the other hand, this numerical example validates that the maintenance model based on the proposed state diagram represents the real world maintenance situation and can be solved using standard Markov methods to get accurate results. The main drawback of the proposed state diagram is the increased number of states. If the classical state diagram has n deterioration states the proposed model has (n-2)(n+1)/2 additional states. However, after finding the state probabilities, the number of states of the proposed state diagram can be again reduced to n. The reduced version of the proposed state diagram is shown in Fig. 3 and it is very similar to the classical state diagram shown in Fig. 1. The only difference between the reduced version of the proposed state diagram and the classical state diagram is the two transition rates from S2 to I2 and from S3 to I3. For the reduced version of the proposed state diagram, 2-new and 3-new can be found using frequency balancing techniques. The expressions for 2-new and 3-new are given in (1) and (2).

TABLE III VISIT FREQUENCIES [1/YEARS] State S1 S2 S3 I1 M2 I2 M3 I3 F Classical model 0.6080 0.2844 0.0639 0.3663 0.2204 0.2204 0.0426 0.0426 0.0213 Proposed model 0.5143 0.2486 0.0850 0.3098 0.1637 0.1637 0.0442 0.0442 0.0408 TABLE IV MEAN DURATIONS [YEARS] State S1 S2 S3 I1 M2 I2 M3 I3 F Classical model 1.2048 0.7752 0.6667 0.0028 0.0028 0.0028 0.0056 0.0028 0.0833 Proposed model 1.2048 1.1787 0.9611 0.0028 0.0028 0.0028 0.0056 0.0028 0.0833 TABLE V Monte Carlo simulation [8] 1.200 1.185 0.958 0.003 0.003 0.003 0.006 0.003 0.083 Monte Carlo simulation [8] 0.514 0.250 0.084 0.308 0.166 0.166 0.044 0.044 0.041

2-new =

" P2' 1 +P2 2 P2

(1)

" Where P2 =P2' +P2 and P2' , P2" and P2 are the state

probabilities of the states S'2 , S" and S2 respectively. 2 3-new = P3' 1 +P3" 2 P3 (2 )

Where P3 =P3' +P3" and P3' , P3" and P3 are the state
' probability of the states S3 , S" and S3 respectively. 3

302

be solved using Markov techniques to obtain valid results.

1
2 3
2-new

3-new

Fig. 3. The reduced version of the proposed state diagram

Finally, this paper discusses the main drawback of the proposed state diagram due to the increase in number of stages. A simple technique to reduce number of states in the proposed state diagram is presented. This can be used to obtain accurate results while reducing the complexity in further steps of maintenance modeling. The idea behind this proposed state diagram can be used to convert existing maintenance models based on state diagrams into better maintenance models, which represent the real maintenance situation and provide accurate results. REFERENCES
[1] [2] J. Endrenyi, Reliability Modeling in Electric Power Systems. Chichester, U.K.: Wiley, 1978. G. J. Anders, J. Endrenyi, G. Ford, and G. Stone, A probabilistic model for evaluating the remaining life of electrical insulation in rotating machines, IEEE Trans. Energy Convers., vol. 5, no. 4, pp. 761767, Dec. 1990. G. Anders, J. Endrenyi, G. Ford, J. Lyles, H. Sedding, J. Maksymiuk, J. Stein, and D. Loberg, Maintenance planning based on probabilistic modeling of aging in rotating machines, in Proc. 1992 CIGRE Int. Conf. Large High Voltage Electric Systems. J. Endrenyi, G. Anders, and A. Leite da Silva, Probabilistic evaluation of the effect of maintenance on reliabilityAn application, IEEE Trans. Power Syst., vol. 13, no. 2, pp. 576582, May 1998. G. J. Anders, J. Endrenyi, and C. Yung, Risk-based planer for asset management, IEEE Comput. Appl. Power, vol. 14, no. 4, pp. 2026, Oct. 2001. G. J. Anders and J. Endrenyi, Using life curves in the management of equipment maintenance, in Proc. 2002 Probabilistic Methods Applied to Power Systems (PMAPS) Conf., 2002. P. Jirutitijaroen and C. Singh, The effect of transformer maintenance parameters on reliability and cost: A probabilistic model, Elect. Power Syst. Res., vol. 72, no. 3, pp. 213224, 2004. Thomas M. Welte, Using State Diagrams for Modeling Maintenance of Deteriorating Systems, IEEE Trans. Power Syst., vol. 24, no. 1, pp. 58 66, Feb. 2009. C. Valdez-Flores and R. M. Feldman, Survey of preventive maintenance models for stochastically deteriorating single-unit systems, Naval Res. Logist., vol. 36, no. 4, pp. 419446, 1989. M. Rausand and A. Hyland, System Reliability Theory: Models, Statistical Methods, and Applications. Hoboken, NJ: Wiley-Interscience, 2004. M. Stopczyk, B. Sakowicz and G.J. Anders, Application of a semiMarkov model and a simulated annealing algorithm for the selection of an optimal maintenance policy for power equipment, Int. J. Reliability and Safety, Vol. 2, Nos. 1/2, 2008. Singh. C. and Billinton. R., System Reliability Modeling and Evaluation. London, UK: Hutchins on Educational Publishers, 1977. Asgarpoor S. and Mathine M.J., Reliability evaluation of distribution systems with nonexponential down times, IEEE Trans. Power Syst., vol. 12, no. 2, pp. 579-584, , 1997. J.V. Bukowski, "Using markov models to compute probability of failed dangerous when repair times are not exponentially distributed", in Proc.2006 Annual Reliability and Maintainability Symposium (RAMS '06), pp.273-277.

This reduced version of the proposed state diagram can be used for maintenance modeling to get the same accurate results with a reduced complexity. IV. CONCLUSION

There are several advantages of using state diagrams in maintenance modeling and these models provide accurate results when the inspection rates are periodic. However, classical maintenance models based on state diagrams does not represent the maintenance situation in the real world, when inspection rates are non periodic. Therefore, the focus of this paper is on maintenance models based on state diagrams with non-periodic inspection rates. According to classical state diagrams with non-periodic inspection rates, when the system is deteriorated next inspection rate is renewed. However in reality, inspection rate is changed, only after an inspection or a maintenance action, which provides the knowledge about the deterioration state of the system to the system operator. Therefore classical state diagrams do not represent the real situation, and in this paper, poor model properties of classical state diagrams are noticed. Then, by introducing some new states to the classical state diagram, a new state diagram is proposed to represent the maintenance situation in the real world. In this proposed state diagram, the next inspection rate is changed only after the operator gets knowledge about the current deterioration state of the system by conducting inspection or maintenance. Although there are state changes due to deterioration unknown to the operator, the inspection rate is not changed prior to an inspection. There are two methods to calculate reliability measures of the maintenance models. If the maintenance model is based on a state diagram standard Markov methods are used, otherwise Monte Carlo simulation is used. This paper applies these two methods to a numerical example. In this numerical example both the classical state diagram and the proposed state diagram are realized using standard Markov methods to obtain state probabilities, visit frequencies, mean durations, MTBF and MTTFF. These results are compared with the results obtained using Monte Carlo simulation, which has been carried out with the understanding of the real maintenance situation. Although, the classical state diagram provides inaccurate results, the proposed state diagram provides the same results that are obtained using Monte Carlo simulation. Based on the results, it can be concluded that the proposed maintenance model represents the maintenance situation in the real world and can

[3]

[4]

[5]

[6]

[7]

[8]

[9]

[10]

[11]

[12] [13]

[14]

303

Das könnte Ihnen auch gefallen