Optimal cording of an artificial neural network in impact-echo testing

摘要

Impact-echo testing instrument designed by artificial neural network is presented Impact-echo signals of bottles with and without cracks on the mouth are transformed into frequency spectra, and their intensities are compressed by considering a human's hearing character The identifying result at the compression coefficient of 0.6 attains 98.5 percent correctly. It is observed that the judgements of cracks in the neural network are made by the relatively low intensity signals.