Spectral Characterization of Pulsed Ultrasound Using Neural Networks

摘要

A novel nondestructive evaluation technique that uses the spectral signature of apulsed ultrasound signal to identify metals had recently been abandoned because of the difficulty in interpreting the results. Traditional analysis is inconvienent to apply to this type of problem because of the complicated, noisy and incomplete nature of the data. Neural networks provide a radically different approach to computation. These massively parallel systems provide a mechanism to extract pertinent information from input data while maintaining a high degree of fault tolerance. This report discusses design of a neural network system capable of accepting data from nondestructive test equipment and producing output relative to the quality of the sample being tested.