One method for condition assessment of electrical equipment

After opening up the electricity markets to more competition the transmission
system operators experience an increasing pressure of costs. Hence they are
forced to make use of all possible economy measures. In order to reduce the
costs in the field of maintenance activities the transmission system operators
establish an asset management system. Software support is essential when
setting up an asset management system. There is an integrated system for
maintenance scheduling which optimizes the point in time when necessary
maintenance actions have to be carried out considering system specific
constraints. This system uses the type of maintenance actions, their duration
and the allowed maintenance period as input data which are determined on the
basis of intelligent condition assessment of the equipment. For the purpose of
condition assessment substantial input data is necessary and available. This
thesis presents an approach to condition assessment based on Dempster-Shafers
theory of evidence which can be regarded as generalization of Bayesian theory
of probability. Different possible diagnoses are modeled by Markov trees and
it is shown how to propagate evidence using Markov trees.

By way of example a power transformer is used to show the application of the
system. Power transformers are expensive devices outfitted with substantial
protection and monitoring equipment. The qualitative relationship between
primary information and possible diagnoses is modeled by a Markov tree. The
quantitative relationship is modeled by basic probability assignments which
map the value of the input data to mass numbers used by theory of evidence.
After processing all input data the correct diagnosis can be determined. The
allowed maintenance period is calculated by extrapolation of the
characteristics of the degree of belief as a function of time. An extension of
the system considers the age of primary information. Recent information are
given a greater weighting than older information. As a result it is possible
to detect a lack of primary information and cause e.g. visual inspections to
be done. Finally a complete model of a power transformer is given. In order to
verify the system the results of dissolved gas analyses of three different
power transformers are used to determine the correct one of six possible
diagnoses.

Keywords

electric power system, transmission network, system operator,
asset management, maintenance, theory of evidence, Markov tree,
condition assessment, power transformer, dissolved gas analysis
