A neural network architecture for ultrasonic nondestructive testing

This paper considers the application of neural networks for detecting defects in materials consisting of dense micro-structures using ultrasonic pulse-echo systems. The motivation for this work is the desire to detect defects modeled by small complex scattering centers. To preserve signal features useful for defect detection, minimal preprocessing is performed on the data presented to the neural network. This paper describes modifications to the typical feed-forward multi-layer perceptron architecture for direct application to sampled RF ultrasonic A-scans.