9522266 Cheng The proposed research is focused on, on-line dielectric diagnostics of operating transformers in the framework of electric PD detection. The proposed approach consists of three fundamental parts: (1)PD signal detection, (2)Signal preprocessing and (3)classification. In the first part, temporal PD signals as well as external noise signals coupled through power lines are collected by using a data acquisition system developed previously. These signals are picked up from ground wires of both transformer and bushing, from the air, or others. Then they are fed into the block of signal preprocessing, where noise cancellation and other classical techniques of signal processing are employed. The study of stochastic characteristics of PD's and noises will help to design reliable and effective filters. Features used in the PD recognition block will be determined in the task of PD & noise characterization. After performing feature mapping from the filtered temporal signal, a neural network is used to identify defect types of dielectric insulation in operating transformers. The neural network is trained with the knowledge of PD patterns from the task of PD characterization. ***
