| A transformer is one of the major components in any electrical network. Losing a transformer can cause serious outages to the power being delivered to customers. Knowing that, it is of utmost importance to test the transformer and to make sure it can survive during its life-cycle starting from its manufacturing stage and ending in its failure/replacement stage. One of the transformer tests is the Sweep Frequency Response Analysis (SFRA) test. SFRA has become a powerful technique since it is a non-destructive test. It consists of injecting a sinusoidal signal, with constant voltage and with sweeping frequency (20Hz-2MHz), to one of the transformer's terminals and measuring the response at the neutral terminal. It is used to check the integrity of the transformer, mainly its winding and core parts.;In presence of noise, SFRA is heavily affected and its waveform gets distorted. In such a case, a healthy transformer may be misinterpreted and can lead to a wrong conclusion. Therefore, noise removal through the use of a signal processing technique is important to avoid misinterpretation and wrong conclusion. Moreover, in the analysis of SFRA, some researchers have been using a cross correlation coefficient as a measure to decide whether this transformer is healthy or not. However, there is no standard for this cross correlation coefficient and if we have a faulty transformer, its cross correlation coefficient might show high value (above 90%).;Due to the above reasons, the Time-Frequency Analysis (TFA) is proposed rather than the classical SFRA. In the presence of noise, TFA will be less affected since the noise will spread all over the joint 2D time-frequency plan. Moreover, a 2D cross correlation coefficient, as a measure of accuracy for our technique, is proposed instead of the 1D cross correlation coefficient that is used for the classical SFRA. The 2D cross correlation coefficient can recognize the type of faults much better than the 1D cross correlation coefficient. This is because TFA is more sensitive than the classical SFRA; consequently, the cross correlation coefficient will give better discrimination of the fault.;Furthermore, after the fault is detected inside the transformer, it is very important to know the cause of the fault which can be a core problem or winding problem. Therefore, modelling the transformer is necessary. In this work, a new technique for transformer modelling that can be used for SFRA is proposed. Many models have been proposed by many researchers, which are either too complicated or they lack the matching idea between the actual curve and the modelled one. The proposed model is simple and it closely matches the actual curve. This model will help us in interpreting the faults happening inside the transformer in terms of its RLC values. |
