Bettayeb F., Boussiha K., Benachir D.

The precise detection of echo-graphic pulses embedded in background noise is a problem of major importance in ultrasonic non-destructive evaluation of materials. Structural noise is a significant constraint to the visibility of flaw echoes in ultrasonic evaluation of highly scattering materials. In order to enhance the defect to background noise, several signal processing methods have been proposed, including split spectrum processing, time frequency distributions, and wavelet transform. All this methods give an output SNR improvement when this ratio is high. In the case of ultrasonic testing, the signal energy coming from the defect fills lower frequencies than the one coming from the material structure, due to dispersion phenomena and the non stationary nature of the ultrasonic travelling wave. We try in this work to extract the noise features by finding its analysing wavelet function. And we propose a procedure based on an energetic threshold of the ultrasonic signal, then a smoothing of the energetic parameters of the noise function. This algorithm has given a good filtering of the ultrasonic signal as well as the detection of little defects in very noisy signals. The experiments have been conducted on welded pieces and on artificial flaws.