Research On Statistical Lifetime Modelling Of Transformer

As one of the most capital-intensitve assets in a substation,power transformers are key components in power networks which are to deliver energy and transform voltage.To ensure the reliable operation of power system,effective transformer asset management is essential to maximally prevent the occurrence of unexpected failures of transformers.Replacement volume prediction for an ever-ageing transformer fleet in the mid-and long-range is one of the core elements for power transformer asset management.Transformer lifetime model,representing the statistical distribution characteristics of transformer failure times,is the key to reach a reasonable prediction.The development of transformer lifetime model,however,is always hindered by the lack of failure data and the non-exhistence of univerally applicable empirical model.It is therefore of great significance to analyse the effect of statistical data on transformer lifetime modelling which would be helpful for an appropriateness evaluation of a transformer lifetime model.For the case that transformer failure data is limited which prevent the proper development of lifetime model,experts' judgement on the distribution of transformer lifetime models are firstly summarized,based on which an approach of Bayesian Updating is proposed to modify the parameters of lifetime models.Case studies show that by adopting the proposed method transformer lifetime model is able to incorporate expert judgement with field failure data so that the local situation for a specific transformer fleet can be reflected.The proposed method also enables a sequential updating approach,whenever new field failure data is collected,the existing model can be improved in a consistent and progressive manner.For the case when appropriate amount of of failure data and censored data can be collected,a Monte-Carlo simulation approach for generating field collected transformer lifetime data is firstly proposed based on the two-parameter Weibull distribution.Six commonly used Weibull parameter estimation methods are then compared based on the simulated lifetime data.The maximum likelihood estimation method is found to be the optimal one among the six which provides the most accurate estimation results.The method is therefore chosen to estimate the Weibull parameters in the following simulation studies.The effects of data quality on the accuracy level of estimated Weibull parameters are then thoroughly analyzed,including the effect of censored data have on the lifetime models as well as the influences that the sample sizes and censoring rate have on the accuracy of the results.It is found that the relative root mean sqaure error of the estimated parameters can be reduced by either increasing the sample size or decreasing the censoring rate or both.A minimum number of 20 ageing-related failures is expected to ensure the desired accuracy level of relative root mean sqaure error lower than 0.25.To deal with the problem that ageing-related failure cannot be effectively distinguished from random failure relying entirely on pure statistical analysis which greatly impacts the accuracy of derived transformer lifetime model,the formation of random failure mode and ageing-related failure mode as well as their corresponding statistical characteristics are analyzed.Postmortem analysis is firstly proposed to distinguish the two failure modes,based on which random failure model and ageing-related failure model are suggested to be built separately.An approach of combining the random failure model and ageing-related failure model is then proposed.Results of the case study show that ageing-related failure mode can be effectively identified by adopting the proposed approach.The accuracy of derived combined lifetime model is greatly improved as well.For the case that condition information as well as a small amount of failure data can be collected,the degradation process of power transformer is firstly analyzed.The mathematical representation of the degradation process is studied,based on which a two-step data sampling and data simulation approach is proposed.Distributions of key state transition times are firstly modelled based on which and considering the physics of degradation process,a Monte-Carlo simulation framework is proposed to simulate large amount of state transition data as basis for the derivation of conditional hazard rate model.Results of the case study show that the sample size of failure data is effectively enlarged through introducing the artificial failure threshold which therefore improves the derived lifetime model.Remaining lifetimes obtained from the derived conditional hazard rate models for the case study are compared the corresponding results obtained from EPRI model which proves the applicability and effectiveness of the proposed method.The consistency of mean lifetimes derived from the case study results,EPRI model as well as the previously derived combined lifetime model indicates that it is appropriate to consider the mean lifetime of transformer as around 60 years before more failure data are collected.