Combining Artificial Neural Network for diagnosing polluted insulators

This paper presents a method to classify the current polluted level on insulator surfaces, i.e., to diagnose the operational conditions of the electrical system isolation through pattern recognition techniques using the ultrasonic signals obtained from surface discharges on outdoor insulators. Pattern extraction techniques on the input signals by Artificial Neural Networks were used in order to enable a reliable computation during the training. It can be point out that the area centroid of the ultrasonic signals showed a powerful extraction technique. Here, the Multilayer Perceptron Network was used as a single classifier or as a combination of multiple classifiers. Moreover, the developed networks have one or six neurons in their output layer to represent the classes of pollution. A comparison among the four developed neural net models shows the improvement of the networks with six output neurons and that the use of combined models is a powerful technique for this type of application.