Ultrasonic Wave Feature Extraction for Neural Network Classification in not Accessible Pipes

The ultrasonic inspection technique can be very useful to determine the state of non reachable structure. In this paper a method based on the neural network classification to evaluate the corrosion level of non accessible pipes is shown. A set of optimal features constitutes the database and feeds the neural network. These features are chosen by time and frequency features extracted from simulated ultrasonic waves. The results show that the method perform a good recognition rate and the different classes are useful for the human decision to evaluate the corrosion level of the pipe under test. Key-Words: Ultrasonic waves, Neural networks, k-nearest neighbour, non accessible pipes, feature extraction, non destructive evaluation.