Fault Diagnosis System for Power Transformers

Abstract. This paper introduces an artificial neural network (ANN)based fault diagnosis system (FDS) for power transformers. Thesystem is designed to detect, localize and finally classify faults. Theproposed FDS consists of three hierarchical levels. In the first level, apreprocessing procedure for input data is performed. In the secondlevel, an ANN is designed to detect the fault and localize its side. Inthe third level, there are two sub-diagnosis systems. Each system isdedicated to one side and consists of one ANN designed to classifythe fault. This ANN is also cascaded with four parallel ANN s utilizedto identify the faulted phase. The performance of FDS is evaluatedusing samples from local measurements (three-phase primary voltageand primary secondary currents). These samples were generatedusing the EMTP simulation of the High-Dam 15.75/500 kVtransformer substation in the 500 kV Upper Egypt network. Differentfault types were simulated. Fault location and incipience time werealso considered. Evaluation results proved that the performance of theproposed FDS is promising.