Supervised backpropagation neural networks for the classification of ultrasonic signals from fiber microcracking in metal matrix composites

The results of the application of supervised backpropagation neural networks to the classification of ultrasonic signals obtained from a model metal matrix composite are presented. This composite is made of a single fiber embedded in a Ti-24Al-llNb matrix and is used for the characterization of the fiber-matrix interface. The neural network is implemented in the frequency domain with two hidden layers and shows excellent discrimination capability when the network is trained with a judicious choice for the training set of ultrasonic signals. The sensitivity of the performance of the network to the number of examples used for training and the robustness of the algorithm to the change in the training set are discussed