Reliability estimation for populations with limited and heavily censored failure information Lukasz Chmura, P.H.F. Morshuis, E.Gulski, J.J. Smit Delft University of Technology Delft, The Netherlands L.a.chmura@tudelft.nl Anton Janssen Liander – The Network Operator Arnhem, The Netherlands Abstract—Statistical analysis of the life data, is a useful tool helping to assess the life-time of populations of high-voltage components. More specific, the results of such analysis give overview over the failure behavior of the population under investigation, i.e. number and trend of expected failures. For the analysis, the detailed information about ages and numbers and ages of installed units and failed units has to be collected. Subsequently, the distribution representing the behavior of the population is fitted to the data. The latter allows deriving the time-dependent failure rate function, which in turn, directly indicates the trends of the future failures. However, this method requires homogeneous and independent data of sufficient amount. The latter becomes a problem, particularly that for past periods the failure data is often unavailable. It is important to estimate the population reliability and number of expected failures, for the whole population of components being operated. This is also important in the case when the available failure data comes only from one part of the area where the components are installed. In this paper we will show how to deal with populations where the available failure data is heavily censored, and what will the influence of the data division according to the regions in which the transformers are operated, on the failure expectancy. Keywords-failure analysis, transformers, tap-changers, data censoring, reliability I. INTRODUCTION The reliability of transformers is crucial for the reliability of the transmission and distribution network. For that reason it is necessary to assess the life-time of the population of the transformers, so that the replacement and spare units policies can be planned in advance. The mentioned policies aim to assure redundancy of the whole network, if certain component of the network fails. This situation refers to the N-1 criterion, which says that if any single component of the power network fails, the other components should be able to take over the load, so that no single user is disconnected from the supply. The statistical analysis is often used as one of the methodologies for the life-time assessment of the high-voltage components [2]. Fitting mathematical model to the life-time data of the population and derivation of the describing functions, allows to conclude about the actual life-stage of the population. In addition, the number of upcoming failures can be estimated with given confidence interval However, for the proper statistical analysis, the supplied data must display the following features: homogeneity, independence, randomness and sufficient amount. Firstly, the data is to be drawn from one population , where all factors are constant for the whole period of population operation. Secondly, the data are independent for different subject, and every outcome (failure of the component) is equally likely to occur within the population. Finally, sufficient amount of data has to be provided to enable the proper and confident distribution fitting. Figure 1. Schematic showing the data handling for the statistical analysis For the statistical analysis, the detailed information about ages and numbers installed and failed transformers has to be collected. To ensure the robustness of the analysis, the reason 159 2013 Electrical Insulation Conference, Ottawa, Ontario, Canada, 2 to 5 June 2013 978-978-1-4673-4744-0/13/$31.00 ©2013 IEEE