The use of artificial neural networks for condition monitoring of electrical power transformers

Condition monitoring of electrical plant represents an area of great intere st to both manufacturing and utility companies within the electricity suppl y industry. De-regulation and privatisation entail that utilities must oper ate their systems in an optimal fashion and one of the technologies which c an facilitate this is condition monitoring. Condition monitoring has a numb er of important benefits: unexpected failures can be avoided through the po ssession of quality information relating to the on-line condition of the pl ant and the consequent ability to identify faults or problems while still i n the incipient phases of development; maintenance programmes can be condit ion based rather than periodically based; the plant may be utilised more op timally through the use of information relating to the plant's real-time co ndition and/or performance - for example, the plant may be driven temporari ly beyond its stated capacity if it is known that this will not cause any s hort-term problems. This paper will cover the generic capabilities of artif icial neural networks, in both estimation and classification mode, for cond ition monitoring applications, using examples based around work that the au thors have carried out with respect to the monitoring of a power transforme r. (C) 1998 Elsevier Science B.V. All rights reserved.