Fault Detection Of Substation Power Transformer Based On Sound Characteristics

Transformer fault-detection is the process of monitoring transformers' operating status;when failure occurs,this will allow workers to do maintenance in time,and will even predict the transformers' working status over a period of time.Traditional transformer fault-detection includes condition-based maintenance,routine maintenance etc and requires experienced workers to operate on site regularly,thus lacks timeliness and has certain security risks.Along with the technological progress and industrialization process,there is a surge of electricity consumption and power grid expansion.Traditional fault-detection methods can no longer meet the developing demands.With the improvement of computer and electronic technology,real-time on-line fault diagnosis technology has emerged.The transformer fault-detection method proposed in this paper is based on audio feature analysis.By analyzing and extracting the spectrum characteristics of the sound of transformers,combined with the corresponding detection algorithm,this method can achieve the purpose of transformer automatic fault-detection.Fault-detection technology based on audio signals has been developing for decades.In 1960s,it is mainly used in cutting-edge industries like nuclear power and aviation;in 1970s,it expands to shipping,petrochemical engineering and metallurgy;in 1980s,it further expands into all kinds of industries.When doing maintenance,experienced workers can determine whether failure occurs by listening to the sound of transformers.Based on this fault-detection technique,a fault-detection method in simulation of human hearing is proposed in this paper.Using this method,a variety of transformers' sound samples were collected,thus an audio sample library was established.Through the analysis,study and counting of these samples,a corresponding algorithm was designed,and then a fault-detection system was developed.A feature extraction method based on the spectrum characteristics of transformers;audio data is proposed in this paper.The characteristics extracted from each data was formed into a corresponding one-dimensional vector or two-dimensional matrix.Then PCA and 2DPCA algorithm were used to realize dimensionality reduction and feature extraction.Then SVM algorithm was used to classify audio signals,so as to monitor the transformers'operating status.