Abstract
he traditional maintenance activities in the power transmission and distribution system are usually based on post-accident repairs and periodic maintenance. Thus the maintenance personnel cannot immediately know the health status of transformers and take appropriate reactive measures. Monitoring the status of transformers passively can cause irreparable equipment destruction or huge financial loss. In view of this, this research uses the real-time monitoring equipment to collect the empirical data of operational transformers of Taiwan and Australia and analyzes the impact of the overall health status of transformers. The dataset of transformer from Taiwan and Australia contains much information, such as quality of the oil, the oil temperature, dissolved gas contents, furfural contents, electrical appliances and equipment position. However, this research focuses on the transformer failures which results from electrical stress and thermal stress. We extract nine kinds of dissolved gas contents (i.e. O2, H2, N2, CO2, CO, CH4, C2H6, C2H4 and C2H2) to diagnose the health status of transformers and identify the key factors which affect the health status of transformers by Principal component analysis. In addition, this research develops a fault diagnosis platform and embeds Principal Component Analysis (PCA) and Back-Propagation Artificial Neural Network (BP-ANN) into the platform. The key factors extracted by PCA are put into the fault diagnosis platform as input and a fault diagnosis model is established. The fault diagnosis model can help equipment owners, maintenance personnel or relevant units acquire the information of transformers immediately. The instant information provides adequate and flexible time for relevant units to develop the most appropriate maintenance strategy which can reduce the probability of transformer failure and extra cost by a serious fault.