Time-varying prediction filter for structural noise reduction in ultrasonic NDE.

Predominant physical phenomenon in highly scattering materials is the attenuation due to dispersion. Therefore, received echo has high frequencies more severely attenuated than low frequencies and the structural noise can be modeled as a non-stationary random process. Most of the proposed techniques for enhancing the flaw visibility do not exploit the frequency dependency of the incoming flaw signal, assuming homogeneous behaviour of the insonified material. In this work, a new technique based on exploiting the non-stationary nature of the incoming UT signal is presented. Proposed technique is based on the prediction error obtained with a linear and time-varying parametric model of the noise. By this method, when the analyzed UT echo has only structural noise, the prediction error is low, however, if it contains a flaw, high prediction error occurs because a flaw is a non-predictable alteration of the material structure. Experiments with stainless steel show that this method has an excellent performance on SNR enhancement.