A novel approach for fault diagnosis of power transformers based on extracting invariant moments

This paper presents a new technique for diagnosis of power transformer faults, which have been previously detected. This technique is based on image analysis for identification of different fault types in power transformers, such as single line to ground, double line to ground, line to line, and symmetrical fault. Also, the technique can clearly identify the index of the faulty turn that occurs on the winding in the case of turn to turn and turn to earth. The technique depends on building up the locus of the current at the supply side with the difference between the voltage signals at the primary and the secondary sides of the power transformer. For each locus of finite faulty case, seven features are extracted and stored. These features are based on the moment invariant technique, which is insensitive to translation, rotation, and scale changes. The proposed technique is evaluated via an extensive simulation study for a 132/11-KV power transformer using alternative transient program (ATP). The obtained results verify accurate and fast operation for identifying the type of fault and its location on the faulty phase.