Experience with distribution transformer loading and failure prediction

Abstract

Summary form only given. Utilities have used for years indicators such as loading, hot spot and top oil temperatures to determine ratings and health of substation transformers where the required data is readily available. However, the same level of data is not available for low voltage distribution transformers and loading prediction is difficult. Overload failures of distribution transformers, while not particularly significant to the utility network, can strain maintenance budgets, drain resources and sour customer relationships. Typically utilities use a transformer load management program that aggregates customer usage, estimates a peak and aggregates the peaks to a transformer for determining transformer load. Transformer replacement programs will then use loading as an indicator of replacement priority. However, most utilities have many more overloaded distribution transformers than they can economically replace, and predicting transformer failure is a very inexact science. One cause among several is poor data quality. Yet estimating loading requires good customer to transformer connectivity data and validated usage data at a minimum. We presented here one utility's experience with a continuous program of proactive distribution transformer replacements based on continuous calculation of loading from end use data. This program was the result of an alarming rate of transformer failures experience during several recent hot summers. Byproducts of this program were the recognition that loading alone is not a good indicator of potential outage, and that transformer age appears to have little or no correlation with transformer outages. Three consecutive years of locating and proactively replacing transformers that are likely to cause a transformer outage were presented including: data sources; methods; analysis; results to date; issues; conclusions to date. Calculated loading on actually failed distribution transformers was compared with loading of a large control group of not, or not-yet failed, distribution transformers, showing a clear correlation of calculated loading with probability of transformer failure.