Ultrasonic quantitative flaw evaluation using neural networks

A correction method is presented to eliminate echo amplitude deviation from insufficient coupling andinstrument factor in ultrasonic evaluation of a homegeneous solid.The flaws are then be evaluated by neural networkson using the maximum peak-peak values of the flaw echoes and bottom echoes as characteristic features.
In order to simulate two essential elements of nature flaw-smooth surface and plane with sharp edge,18 representative flaw samples with traverse cylindrical cavity and flat-bottom hole respectively are made.The maximum peak-peakvalues of the echoes of the samples are measured four times,and corrections are made to the data.
The experiment results show that flawes can be classified correctly and sized well.Finally,the errors of themeasurement and the flaw size estimation are analysed.