A Method of Abnormal States Detection Based on Adaptive Extraction of Transformer Vibro-Acoustic Signals

Abstract

State monitoring is very important for the safe operation of high-voltage transformers. A non-contact vibro-acoustic detection method based on the Blind Source Separation (BSS) was proposed in this paper to promote the development of transformer on-line monitoring technology. Firstly, the algorithm of Sparse Component Analysis (SCA) was applied for the adaptive extraction of vibro-acoustic signals, which utilizes the sorted local maximum values of the potential function. Then, the operating states of the transformer were detected by analyzing the vibro-acoustic signal eigenvectors. Different conditions including running normally, increasing of transformer vibro-acoustic amplitude and changing of frequency component of transformer vibro-acoustic were simulated. Moreover, experiments were carried out in a 220 kV substation. The research results show that the number of mixed noise sources can be estimated and the transformer vibro-acoustic signal was always ranked first in the separation signals. The source signals were effectively separated from the mixed signals while all of the correlation coefficients are more than 0.98 and the quadratic residuals are less than −32 dB. As for the experiments, the vibro-acoustic signal was separated out successfully from two voice signals and two interference signals. The acoustic signal reflection is considered as the main cause of the signal interference, and the transformer volume source model is considered as the main reason of unstable vibro-acoustic signal amplitude. Finally, the simulated abnormal states of the transformer were well recognized and the state of the tested transformer was judged to be normal.