Application of stationary wavelet transforms to ultrasonic crack detection

Ultrasonic-based pulse-echo technique has been widely used for non-destructive crack detection. The noisy ultrasonic signals reflected by inhomogeneous materials and other effects add difficulty to pulse echo extraction. In order to address this issue, a method is proposed in this paper to remove noise. Firstly, the raw signal is processed using stationary wavelet transform. Secondly, kurtosis is employed as a criterion to retain the appropriate wavelet coefficients on a specific scale, and to zeroize all wavelet coefficients on other scales. Thirdly, the remaining wavelet coefficients are shrunken by a soft-threshold rule using universal threshold sigma root 2 log(n), where sigma is the maximum standard deviation of wavelet coefficients on all scales before being zeroized and n is the data length. Fourthly, cross-correlation analysis between the de-noised signal and the transmitted pulse signal is conducted. Finally, the time-of-flight of the pulse in a material is measured and the flight distance is calculated. Experimental results indicate that the proposed method can detect the crack position effectively.