Data mining applied to transformer oil analysis data

Abstract

Analysis of oil samples is a standard technique in the electricity industry for monitoring the condition of oil filled plant. Samples are typically taken annually, with more frequent sampling where there is a possible problem. Analyses performed on the oil include: dissolved gas analysis (DGA), colour, moisture level, acidity, breakdown voltage and Furfuraldehyde (FFA) content. In DGA the gases usually considered are: hydrogen, methane (CH/sub 4/), ethane (C/sub 2/H/sub 6/), ethylene (C/sub 2/H/sub 4/), acetylene (C/sub 2/H/sub 4/), carbon monoxide and carbon dioxide; variations in the levels of individual gases, or ratios of particular gases may indicate a problem with the plant. This situation is complicated by the fact that the levels of dissolved gas measured can be affected by the sampling technique and conditions, the laboratory performing the analysis and the duration of sample storage prior to analysis. The results of oil analysis undertaken by The National Grid Company are recorded in a database as records of gas concentrations (in ppm). These records are currently analysed by conventional methods; the reported exercise used unsupervised neural networks to unearth further information.