Dealing with the lack of loading and overloading data to determine the loss of life of the power transformer insulating paper

A.A. ROMERO; F. HARDER; E. MOMBELLO; R. DIB; G. RATTA

París

Simposio; INTERNATIONAL COUNCIL ON LARGE ELECTRIC SYSTEMS, CIGRE; 2012

CIGRE SESSION 44

In the field of power transformer (PT) management, there are efforts to formulate a reliable methodology to estimate a PT risk index that usually is based on two sub-indices: the conse-quence factor (CF) and the probability of failure (PF). While the first sub-index stands on the fact that every PT will fail and the consequences of such failure might be roughly estimated, the second one essentially depends on the monitoring and assessment of the PT condition. A PF assessing method must provide a quantitative or qualitative statement regarding the actual PT condition and relies on condition monitoring tools like vibration analysis, dissolved gas analysis, degree of polymerization measurement, partial discharges analysis, frequency re-sponse analysis, loading analysis, among others. Some of these tools require special equipment, disconnecting, and even opening the PT. However, a well-known, economic and non-invasive method is to indirectly estimate the ageing of the PT insulating paper by means of the loading analysis. Such analysis allows for assessing the deterioration of cellulose caused exclusively by temperature, but by no other ageing agents. The formulation is founded on a set of equations, first, for the hot-spot temperature (HST) calculation as a function of ambient temperature, PT characteristics and the load at each instant, and then, for the equivalent accumulated ageing estimation as a function of the HST and historical load profile. In many cases of in-service PT, the available load profile data is incomplete. Therefore to calculate the equivalent accumulated insulating paper ageing for the whole period of PT operation, the historical load profile has to be estimated. In this paper a method to estimate the historical load and ambient temperature profiles by means of an artificial neural network and Monte Carlo simulations, to account for the related uncertainties, is presented and applied to a currently in-service PT. Furthermore criteria for choosing a suitable HST model are defined. Finally, the corresponding results for assessed PT are presented