Application of neural networks for transformer fault diagnosis

Power transformer is one of the most important components in a power system. It experiences thermal and electrical stresses during its operation. The insulation system consisting of mineral oil and the insulation paper used in transformer undergoes chemical changes under these stresses and gases are generated. These gases dissolve in oil. The extracted dissolved gases are analysed in the laboratory using gas chromatograph. The fault identifications in a transformer are based on certain key- gas ratios. International standards such as IEEE and ASTM are in use for fault identification. However, these standards are not able to diagnose the fault under certain conditions. Hence, there is a need to improve the diagnostic accuracy. In this paper an attempt has been made to diagnose the faults in a power transformer using a three level perceptron network. Three types of neural network simulation models are developed using MATLAB™ Software and trained using the IEC TC 10 databases of faulty equipments inspected in service. The outputs of the Neural Network models are compared with the IEEE and ASTM methods. The comparison of the results indicates that the condition assessments offered by the models are capable of predicting the fault with higher success rate than the conventional diagnostic methods.