الفهرس | Only 14 pages are availabe for public view |
Abstract Power transformers are an expensive critical asset in power networks. Most of the transformers currently in service worldwide are approaching or have already exceeded their design life, consequently, power utilities face a significant risk of power transformer failure. Frequency response analysis (FRA) is the most reliable technique for detecting any mechanical deformation within the transformer. Since FRA depends on graphical analysis, its signature interpretation is still a great challenge that requires a highly specialized person to predict the type of fault. This may lead to different interpretation for the same FRA signature by different experts. Moreover, the current FRA technique cannot detect minor winding deformations. In this research different types of transformer faults are simulated using MATLAB by changing some electrical parameters in the high frequency transformer model. This simulation is carried out on three different transformers to investigate the changes in the FRA signature under various faults. Also, the whole frequency band (10 Hz – 1 MHz) is divided into four regions and some statistical parameters are calculated for each region to investigate the change in these parameters under various faults. These statistical parameters are used as input data for an artificial intelligence model for identifying the fault type. The used model is called adaptive neuro-fuzzy inference system (ANFIS). The ANFIS has the advantages of both fuzzy systems which can incorporate the experience knowledge into a set of rules and the neural networks that can adapt their parameters. The proposed ANFIS model is applied on the three simulated transformers to test its ability of fault identification. |