Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

1

Novel Methods for the Condition Based Maintenance of Distribution Networks


Edmund Handschin*, Ingo Jrgens*, Jens Weller** and Markus Zdrallek**

Abstract The condition based maintenance (CBM) of highvoltage power supply systems provides many economic benefits compared to other maintenance strategies, if relevant data for the assessment of the assets condition are available. In order to collect them systematically, uniform checklists have been developed for different types of assets. They are used as input data for an expert system and a mathematical method, which allow the estimation of the assets condition by calculating normalized parameters. Although the basic approaches of the developed methods are quite different, their application on power transformers and circuit breakers lead to very similar results, so that both methods validate each other to a large extent. Part of the mathematical model is the consideration of uncertainty according to imprecise and missing datasets, so that a margin of deviation regarding the condition parameters can be determined. Index Terms condition assessment, condition based maintenance, power system economics, expert system, theory of evidence

I.

INTRODUCTION

HE liberalization of the electricity markets has lead to more competition and the network system operators experience an increasing pressure of costs. In order to reduce the costs in the field of maintenance and renewal activities asset management systems are established, which are based on different strategies in dependency of individual network topologies and voltage levels. In this context the condition based maintenance (CBM) becomes more and more important for the operation of power supply systems. Particularly in high-voltage (HV) levels it increasingly replaces previous time-dependent maintenance strategies whereas for mediumand low-voltage systems furthermore time- and event-based maintenance concepts are used in general. A necessary requirement for a CBM strategy is the exact knowledge of the current technical condition of an asset [1]. Therefore condition defining parameters must be identified at first. Subsequently, the essential input parameters have to be evaluated and weighted. To gain this information online- and offlinemonitoring systems as well as manufacturer instructions and empirical values can be used. As expensive assets like power transformers in high-voltage systems are equipped with
*: **: University of Dortmund, Institute of Energy Systems and Energy Economics, Germany, 44227. RWE Rhein-Ruhr Netservice GmbH, Pulheim, Germany, 50259.

online- or offline-monitoring systems, most assets in lower voltage levels are operated without monitoring. With an increasing amount of objective values a more precise condition assessment is possible. For these purposes standardized checklists for various kinds of assets have been developed, which contain comprehensive information of online- and offline-monitoring systems as well as statistical and equipment-historical data. Furthermore, empirical values and subjective appraisals of the personnel are part of the checklists. The introduction of theses checklists is part of this paper. In principle, the processing of data to estimate the condition of assets can be performed in different kind of ways by heuristic and analytic methods. In this paper, novel methods for the condition assessment of assets are presented. Therefore, expert systems developed at RWE Rhein Ruhr Netservice- as well as mathematical methods developed at the University of Dortmund- are used. The main reason developing two different methods is to determine, if both approaches come to similar results or not. Thus, the verification of both methods is possible. Furthermore, the analysis of different results allows a combination of both methods, so that benefits can be obtained in an optimal way. The expert system is a two-level procedure, which provides an overall evaluation of an electrical asset and a priority regarding a maintenance or repair task in case of a detected technical deficiency. In the first step an overall index for the examined equipment is generated. This index is calculated from numerically displayed evaluations of technical condition parameters, compliance with regulations and the importance of the equipment in the operated grid. In the second step the combination of an identified technical deficiency and the importance of the equipment provide a level for a priority regarding a maintenance task. The system also contains automated routines to summarise accomplished evaluations according to freely definable criteria. As a result various levels of summarisations (e.g. all transformers in a substation, all circuit breakers in a grid etc.) are possible. The developed expert system has been proved and fieldtested in the grid of RWE Rhein-Ruhr. Subsequently, the procedure has also been applied to various external grid operators. The mathematical methods are based on the theory of evidence, which is a generalization of the probability theory [2], [3]. Resting on the results of those analyses, a condition parameter is modelled, which can be used as an objective

2 pointer for the management of maintenance measurements. Furthermore, the estimation of maximum and minimum values of the parameter is part of the model. Both methods have been applied to HV transformers and circuit breakers. The condition results assessed by the mathematical method are compared to the results achieved from the expert system. of the checklist. Thus, the user obtains the automatically calculated overall results regarding overall index, technical condition, priority of a maintenance task, compliance with regulations and importance of the equipment in the operated grid at a glance. In the second part of the checklist (page 2 and following) the actual technical inspection of the equipment is accomplished by probing and evaluating the listed parameters. In the third part (last page) an inspector has the opportunity to evaluate qualitatively, whether formalities and regulations regarding occupational safety, environmental protection and legal requirements are complied. Also the degree of significance of an equipment in an operated grid can be specified.

II. CHECKLISTS Basically, for a reliable condition assessment an objective and extensive database is necessary. Hence, standardized checklists for all kinds of HV assets have been developed at RWE Rhein-Ruhr Netservice. Lists for the following groups of HV equipment are available: 1. 2. 3. 4. 5. transformers, circuit breakers, converters protection, control and communication technology power lines and cables power supply technology (batteries, rectifiers etc.) plants

III. EXPERT SYSTEM By means of substantiated analyses of operational experiences, qualitative and quantitative relationships between input data and asset condition have been developed. In particular, the weighting of input data and its influence on the condition is essential for a standardized technique and is modelled with the help of expert knowledge of the operational staff. The expert system analyses the assessed parameters of the checklists and provides two results. In the first step an overall condition index c(T,C,I) for the examined equipment is calculated by combining the recorded data of the technical inspection (T), the evaluations of compliance with regulations (C) and the importance of the equipment in the operated grid (I). The index is a number in the range of 0 (= in mint condition/optimal) and 100 (= inadequately/deficient) and is primarily used to create a ranking within a group of equipment of similar type (e.g. all circuit breakers in a grid). Thus, this ranking specifies, which equipment has the highest need for action regarding a maintenance or repair task and so establishes a priority list for maintenance and renewal activities. Taken alone, an overall condition index proved unsatisfactory to provide an exhaustive evaluation of an equipments condition. As a result of the mathematical summarisation of all technical parameters it may occur, that an equipment is overall rated in mint condition or sufficient although single parameters of most importance are classified insufficient or in need of repair. Therefore, in a second step of the procedure these identified deficiencies are shown separately for the purpose of initiating selective measures regarding a maintenance or repair task. Combined with the general importance of the equipment in the operated grid a priority for the selective measure is defined furthermore. A schematic representation of the evaluation concept of the expert system is illustrated in figure 1. The system also contains automated routines to summarise accomplished evaluations according to freely definable criteria. As a result various levels of summarisations (e.g. all transformers in a substation, all circuit breakers in a grid, etc.) are possible and can be charted and tabulated. Thus, the

Besides statistical data like manufacturer, type, age or purpose of use, the checklists contain all parameters, which are essential to define the technical condition of an asset. They also provide the opportunity to evaluate the compliance with regulations and the importance of the equipment in the operated grid. The documented values can be clustered in three different types: 1. 2. 3. continuous measured values description in range of values subjective ratings of the personnel

The continuous values are collected by online and offline measurements and are the most precisely entries of the checklists (e.g. short-circuit impedance of a transformer). Some information can not be documented as continuous values but is stored as discrete ranges. This is reasonable if more detailed data is not significant for the assets condition (e.g. observance of critical values). Furthermore, some information can only be estimated qualitatively (e.g. leakage of mechanical equipment). The third part contains subjective ratings of the personnel and is helpful for the collection of operational experiences. Thus, asset-specific attributes, that can not be recorded but can influence its condition, are regarded. Besides the mentioned data further statistical information is collected in the checklists. Some important examples are age, manufacturer, characteristics, purpose of use and ambient conditions. They can also be helpful for the detailed parameterization of a condition model. The checklist itself is divided into three parts. The first part (page 1) contains basic statistical data (asset owner, location, manufacturer, age etc.) and the summarised results of the single parameter appraisals documented in the following parts

3 two-level evaluation procedure overall index

&
compliance with regulations (C)
legal requirements occupational safety environmental protection

parameter 1 parameter 2 parameter 3 parameter ... parameter n

technical condition (T)

importance (I)

&
need for action (techn.)
Figure 1 Evaluation concept of the expert system

priority

obtainable condition representations of substations, asset groups or entire grids provide a decision guidance for an optimal application of a maintenance or renewal budget. Besides the mentioned procedures further additional functions are implemented in the system. For instance a routine is available, which automatically creates all required checklists for an upcoming examination and stores them in a directory. Evaluation results documented in a checklist can be transferred automatically to a standardized written report for each asset. Also the expenditures of time, personnel and material for an evaluation can be calculated. In the summer of 2006 the developed expert system has been proved and field-tested in the grid of RWE Rhein-Ruhr. Besides the question of suitability and practical feasibility further aspects like expenditures or minimum requirements for a condition appraisal were examined. Subsequently, the procedure has also been applied to various external grid operators.

evidence ([2], [3]) compared to the classic probability theory is the different consideration of uncertainty by the use of two probability values. The degree of belief (Bel) describes the level of certainty, that a statement is true or false. The degree of plausibility (Pl) is the level of the maximum plausibility, that a statement could be true or false at all. Furthermore, contradictory hints for a diagnosis can be processed systematically. In the case of condition assessment it is not necessary to estimate, whether individual diagnoses are true or false but if they have an influence of the asset condition itself. Thus, in this approach the following two statements are defined:
A:

asset condition bad


asset condition well

A:

IV. EVIDENCE MODEL In addition to the expert system a second model has been developed in order to analyze its robustness and to verify it. It makes use of a probabilistic method that is based on the theory of evidence. The general applicability on power-supply equipment and its benefits compared to the standard methods has been determined in [4]. In a former study this method has been applied for a dissolved gas analysis as an important criterion of a power transformers condition [5], [6]. The method presented in this paper allows the condition assessment of the whole asset circuit breaker and power transformer. A. Theory of evidence The main property of the Dempster-Shafers theory of

Because of using two degrees for the probability of a statement a separate consideration of information pro and contra a diagnosis is possible. Compared to the usual probability theory a light evidence for the compliance of a statement ( Bel ( A) ) is not equivalent to a strong evidence for the complementary statement ( Bel ( A) ). This interrelation is illustrated by figure 2.

Figure 2 Degree of belief, degree of plausibility

In this contribution, the range of values is defined by (1).


1 Bel ( A) = Pl ( A)

(1)

Pow er Transform er

G eneric

G as analysis
SC-Im p 1

W inding
SC-Im p 3 SC-Im p 4

A ge Relocations

Load

C2H 4 C2H 6

C2H 2 C2H 4

CH4 H2

SC-Im p 2

Figure 3 condition tree of a power transformer model

The parameterization of the probability values is carried out by modelling specific mass numbers m( A) and m( A) , which characterize the quantitative relationship between input data and asset condition. In this case, the grey area in the medium part of figure 2 can be interpreted as the degree of uncertainty. B. Condition tree By means of a condition tree all relevant input data, which is significant for the condition of an asset, is combined with each other [7]. Thus, the qualitative interdependences between input data and asset components can be modelled. It can consist of several components for modelling individual parts of the asset. In this approach, condition trees for circuit breakers (CB) and power transformers (PT) have been developed. Exemplarily, the PT-model is illustrated in figure 3. It is composed of three components, whose combination leads to the estimation of the total condition. Each component is modelled by an individual Markov-tree, where the input data is processed and combined with each other. General values like age, number of relocations and load have an influence on the transformer as a whole but not only on single components. Hence, the model Generic covers condition changes, which are significant for the entire transformer. The model Gas analysis covers six diagnoses as the results of the dissolved gas-analysis [5]. They are combined with each other to calculate the probability values of the whole component model. The third model covers the condition of the transformers winding by using values of the short-circuit impedance measuring. The quantitative relationship between input data and condition assessment is mapped by the parameterization of the mass numbers. They determine whether a statement is supported, rejected or uncertain in dependency of the individual input data. The consideration of uncertainty allows the modelling of measurement tolerances, systematically errors and inconsistent data. The concrete parameterization of the mass numbers is carried out by analyzing technical thresholds (e.g. DIN, EN), operational experiences, statistical evaluations (e.g. life-cycle-models) and feasible closures. Generally, they can be modelled as continuous and discrete magnitudes. They are combined with the rule of Dempster [3]

for each model. C. Condition parameter The main goal of methods characterizing the condition of assets is the derivation of standardized parameters. Thus, an objective comparability of various assets determining priority lists is possible. Furthermore, different methods calculating such a parameter can be compared with each other. Modelling a parameter using the results of the condition tree method requires a consideration of both probability parameters Bel and Pl. On that score the condition parameter is modelled in compliance with (2).
z ( A) = b Pl 2 ( A) + (k Bel ( A)) 2

z ( A) = b Pl 2 ( A) + ( k Bel ( A)) 2
z( A) + z( A) = 100

(2)

This approach allows the usage of both probability values, in which the weighting of Bel and Pl can be varied by the parameter k. By scaling the parameters on a maximum value of 100 the interpretation of the results for the planning of further maintenance or renewal measurements is possible. For this purpose, concrete guidelines in dependency of the condition parameter have been defined. D. Extreme value estimation In practice, not all of the required model data is available because of operational reasons. The most important cause is the need to take the asset out of service in order to collect relevant data. On the one hand, it is more profitable to determine the condition without taking the asset out of service; on the other hand the condition assessment is more exact, if all data is disposable. Nevertheless, it is important to estimate the minimum and maximum value of the condition parameter in case of complete database to schedule the maintenance tasks in an optimal way. Potentially, concrete measurements can already be assigned by analyzing all parameters of an asset group despite using all relevant data. Hence, the introduced method has been extended by extreme value estimation. For that purpose, in a first step the extreme

5 condition parameters are calculated for every model and are combined to the parameter of the whole asset in a second step. That leads to the calculation of the highest and lowest condition parameters, which could be calculated, if all data would be available. By this means, an area of uncertainty as a result of missing values can be estimated, which contains the true value of the parameters z ( A) and z ( A) .

measurements is made. Nevertheless, it has to be analyzed, if the margin of deviation is acceptable or not. In relation to the results of other assets it can be decided, if the collection of further input data is necessary for scheduling concrete maintenance activities. Potentially, definite priority lists already can be generated under consideration of the calculated deviations, so that further inspections are not necessary. Furthermore, it is possible to estimate, whether prescriptive condition limits are preserved or not. B. Priority lists Besides estimating concrete condition parameters it is essential to assess the results of an asset cluster consisting of several units. Analyzing individual results in relation to others is important for the planning of further maintenance and renewal strategies. Hence, priority lists can be created, which contain the condition parameters c(A) and z(A) of all regarded assets in descending order. Assuming a fixed budget the assets with the highest parameters are treated with the highest priority. The following table represents the priority lists for a group of five power transformers T1-T5 using the condition assessment of expert system and evidence model.
TABLE I PRIORITY LISTS FOR A CLUSTER OF FIVE POWER TRANSFORMERS Name T1 T2 T3 T4 T5 c(A) 20.3 14.6 14.3 12.1 10.2 priority (expert system) 1 2 3 4 5 z(A) 50.0 25.0 22.2 17.3 16.4 priority (evidence model) 1 2 3 4 5

V.

EXEMPLARY RESULTS

A. Individual condition assessment The results of a condition assessment for a single HVpower transformer based on the described expert system and the evidence model are shown in figure 4. As the transformer was inspected during operation, the measurement of the short-circuit impedances could not be performed. The full line illustrates the condition parameter c(A) based on the expert model with its value 12.1. The dashed line indicates the condition parameter z(A) with its value 17.3 based on the evidence model exclusively using the available input data for the generic model and the gas analysis. This is the quantity for a bad condition, so that the quantity for a well condition z(A) results in 82.7. For the winding model, where no input data is available, an extreme value analysis was carried out. The minimum value for z(A) is 11.5 and is signified by the lower dark bar. The upper white bar illustrates the minimum value of z(A) (55.3), which can be achieved in the most unfavourable case, if all required data would be available. Thus, the medium grey area describes the margin of deviation. In case of the availability of all data this part would disappear.

100

75

55,3

50

25 11,5 0 PT_211 z(A)_min z(A) uncertainty c(A ) z(A)_min

It can be recognized, that both methods come to the same result concerning the priority of further maintenance activities. Furthermore, most condition parameters are quite similar, so that both methods endorse each other. Certainly, there are significant differences at the concrete condition assessment of transformer T1. The reason for this is the fact, that a lot of input data was not available. This leads to a high level of deviation using the evidence model. It becomes obvious, that the performing of a complement condition assessment not only requires analyzing the condition parameter but also the consideration of uncertainty. The results for the condition assessment of circuit-breakers show a similar conformability as the results of power transformers presented above. The values of the condition parameters approximate each other with increasing availability of input data.

Figure 4 Condition assessment of a power transformer

condition param eter

It is obvious, that both methods come to very similar results concerning the condition assessment in this case. The area of the parameter deviation z(A) imbeds the value of c(A) so that it can be supposed that a uniform prediction regarding further

VI. CONCLUSIONS Especially assets in HV-networks are maintained in dependency of their condition. Basis for the appliance of CBM is an objective condition assessment to perform

6 maintenance activities in an optimal way. For these purposes, standardized checklists have been developed, which contain different types of input data. They are gained by online- and offline-monitoring systems and inspections. The processing of these data is carried out by two different techniques based on expert systems as well as based on mathematical methods. The expert system is parameterized by the extensive level of expert knowledge. At the mathematical system, the quantitative interrelations are modelled by mass numbers, whereas the qualitative relationships between input data and asset condition are mapped by condition trees. By combining both degrees of probability resulting from the theory of evidence two parameters for the characterization of the asset condition have been defined. They include uncertainties resulting from imprecise or inconsistent data. Furthermore, unavailable input data is used for estimating extreme condition parameter values, so that a margin of deviation can be calculated. By means of these results in relation to the conditions of other assets concrete maintenance measures can be performed. In dependency of the margin of deviation it can be decided, whether further data acquisition is necessary or not. In this way the economic scheduling of maintenance measures is supported. The comparison of the results calculated by both methods for a group of power transformers shows a high level of correlation. Although there are strong differences between both models, the main statements concerning further maintenance measurements according to priority lists are unitary. Differences between single condition parameters are grounded in missing input data, so that a high margin of deviation occurs. This has to be regarded for the scheduling of further maintenance tasks. The main result of the comparison of both methods mathematical model developed by the University of Dortmund and expert system of RWE Rhein Ruhr Netservice- is that they almost lead to the same condition assessment. Slight differences in the results are comprehensible and so both methods approve each other. As a next step the implementation of one combined method is scheduled to take advantage of the benefits of both systems. Therefore, an implementation of an IT-tool is planned, which comprises the large level of expert knowledge as well as the mathematical methods of the evidence model. Furthermore, the application on additional types of assets is planned to process the condition assessment of completely power systems. Thus, an optimization of further maintenance activities can be reached due to technical and local dependencies of different assets.

VII. REFERENCES
[1] [2] C. Neumann Monitoring und Diagnose als Werkzeug des Assetmanagements, in: ETG-Fachberichte, Vol. 104 (2006). A. P. Dempster, Upper and lower probabilites induced by a multivalued mapping, Annals of Mathematical Statistics, vol. 38, pp. 325-339, 1967. G. Shafer, A mathematical theory of evidence, Princeton University Press, London: Princeton, 1976. U. Neumann, Integrierte Instandhaltungsplanung fr elektrische Energiebertragungssysteme, VDI-Fortschrittsberichte Reihe 21, Nr. 249, VDI Verlag, 1998. W. Feilhauer, Ein Verfahren zur Zustandsbewertung elektrischer Betriebsmittel, Ph.D. dissertation, Dept. Electrical Eng. and Information Tech., Univ. Dortmund, Germany, 2005. W. Feilhauer and E. Handschin, Interpretation of dissolved gas analysis using Dempster-Shafers theory of evidence, Proc. of 9th International Conference on Probabilistic Methods Applied to Power Systems, Stockholm, Sweden June 11-15, 2006. G. Shafer and P. Shenoy, K. Mellouli, Propagating Belief Functions in Qualitative Markov Trees, International Journal of Approximate Reasoning Vol. 1 (1987), Nr. 4, S. 349-400.

[3] [4]

[5]

[6]

[7]

You might also like