Mathematical Modelling of End-of-Life of Power Transformers in Perspective of System Reliability

Abstract

Asset management strategies are needed by electrical utilities to effectively deal with the growing presence of ageing network infrastructure. Large network assets like the power transformer are approaching their anticipated design lives and have raised concerns over the negative effects of failure (e.g. safety, environmental, technical and commercial). However, the high capital outlay and long lead-time needed to replace large network means that this is a complex, costly and time-consuming process. Appropriate forward planning is, therefore, a prerequisite to ensure the continued success of the modern power system. In order to effectively determine immediate and future network investment strategies a high-level understanding of transformer reliability is required. In this PhD thesis the general methodology on how to create a mathematical transformer failure model is discussed for the UK transmission system operator National Grid. The work conducted focuses on two widely used reliability methodologies; statistical modelling and degradation modelling. Statistical modelling of component lifetime data is conventionally used in reliability studies, however, because of the limited availability of transformer lifetime data and subsequently high censoring rate, its use in transformer reliability studies is limited. Therefore, through a series of Monte Carlo simulations, a study on the data quality requirements for statistical analysis of populations containing different sample sizes and percentages of censored data is conducted. Through these simulations it is subsequently demonstrated that by focussing on estimating characteristic features like percentiles of the lifetime model a more practical means of assessing data quality is achieved. As an alternative approach to modelling transformer reliability it is demonstrated how degradation based transformer reliability models can be developed using either accelerated life test data or transformer scrapping data. In the first instance accelerated life test data is used to consider the combined effects of thermal and mechanical stress on the transformer insulation wear life. The results are used to estimate transformer insulation wear life through the application of a cumulative damage model and a Monte Carlo simulation is subsequently used to assess the short-circuit performance of aged network transformers. From this simulation it is shown that two distinct lifetime regions are identifiable under what are considered to be 'typical' network conditions. Meanwhile, in the second instance, a combination of health index (HI) data and paper insulation measurements (made on 70 scrapped and failed transformers) are used to develop a condition-based degradation model. The developed model allows the possibility of planned (soft) or unplanned (hard) transformer failures to be assessed according to a transformer's assigned HI. Valuably, by utilising scrapping data, it is possible to assess reliability based on past operating experience of similar transformers operating in the same environment. When applied to National Grid's transformer population the initial results suggest that transformers with 'healthy' HI scores are able to maintain a high and near constant level of reliability, however, beyond this point, reliability declines as the condition of the transformer worsens.

Date of Award	1 Aug 2015
Original language	English
Awarding Institution	The University of Manchester
Supervisor	Zhongdong Wang (Supervisor)