Diagnosis Research On Internal Spark Discharge Fault Of Transformer Based On Audio Signal

Transformer internal discharge fault diagnosis based on audio signal is not affected bythe interference of electric field and magnetic field, its monitoring process does not affectthe normal operation of the power transformer. This paper introduces and compares theadvantages and disadvantages of the ultrasonic detection method, noise analysis andvibration detection method, on the basis of the simulated experiment of the transformerspark discharge in the laboratory, the feature extraction of the audio sound of thetransformer spark discharge had been done.The major research work of the paper is on the audio signal of the transformer sparkdischarge which frequency is about20Hz～20kHz. the acoustic mechanism of thetransformer under normal operating conditions is theoretically analyzed, the fault type ofthe transformers are analyzed in detail, The audio signal characteristics corresponding toeach fault type is presented. In allusion to the audio signal characteristics, the detectionsystem is designed, and to study the feature extraction of the non-stationary audio signal,therefore, to develop the Analysis Software of the non-stationary audio signal. The normaloperation of the audio signal is field measured, which serves the groundwork to therecognition of the fault audio signal characteristics. The typical model of the transformerinternal spark discharge fault is built in laboratorial conditions, to analysis the experimentalresults and to build the mathematical model of the audio signal of the transformer internalspark discharge.The outcome of the experiment shows that Time-frequency analysis method andinterval energy extraction from sound signal extraction can reflect the real state of theequipment. The audio signal’s frequency spectrum of the normal operation transformerpresents the line spectrum of100Hz and its frequency-doubled components200,300Hz. The audio signal’s frequency spectrum of the transformer internal spark discharge, whichmainly fluctuates within1000Hz, is presented in the continuous spectrum with envelopcurve.