Transformer assessment using health index – Part I

Transformer assessment using health index – Part I

1.    Introduction

Over the years, transformer health indexing (HI) has become a popular tool for performing transformer health assessments on a larger fleet of transformers. HI is a tool that allows asset engineers / managers to make informed decisions by processing available data of the transformer and convert those into an overall “condition” score. This condition is usually based on “scores” and “weighting”, which are calculated from a set of algorithms designed to evaluate both field conditions, inspection results, on-site test results, etc. Sometimes there are additional sub-algorithms which assess different subsystems of the transformer, and the subsystem tiered values are finally combined to form the final HI, which corresponds to the overall condition of the transformer.

Read the full article in PDF

Transformer health indexing has become a popular tool for performing transformer health assessments on a larger fleet of transformers

The first model was proposed in [1,2]. In this model, HI was developed as a practical tool, which combined results from the laboratory and field tests, field inspections, and general operating conditions. All the data was then combined and converted to a quantitative index which represents the overall health of the transformer. There are three basic requirements to develop this index: inputs, algorithms, and outputs. Inputs can range from quantities, measured regularly as part of routine maintenance by the power transformer owner, such as:

•      Dielectric strength, dissipation factor, acidity, moisture, colour, and interfacial tension of the oil

•      Dissolved gas content of the oil

•      Furans content of the oil

•      Transformer load and age.

Algorithms can range from simple weighted average [3], logarithmic scoring [4], tiered approach [5], group scoring [6], subset indexing [7], fuzzy logic [8], regression neural network [9], etc. Other widely used models include [10] and [11]. Outputs can be constructed in many ways. The classification designed in [1,2] with an expected lifetime is shown in Table 1.

Vol. 7 Issue 4