Transformer-fault diagnosis by integrating field data and standard codes with training enhancible adaptive probabilistic network

A transformer-fault diagnosis system (TFDS) using a probabilistic neural network (PNN) and IEC/Cigre standard codes is proposed. Many artificial neural networks (ANNs) have been proposed before, however, the slow repeated iterative process and poor adaptation capability for structural data restrains the ANN applications. An effective and flexible PNN could overcome these drawbacks. In this paper, a PNN analyses the transformer´s dissolved gas content to identify faults, while using the gas ratios of the oil and cellulosic decomposition to create training examples. Retraining can be done by adding new examples and new hidden nodes for easy adaptation without doing any computed iteration. The commonly used Excel was integrated to provide a convenient man-machine interface. Computer simulations were conducted with diagnostic gas records, to show the effectiveness of the proposed system.