Robust Signal Processing for Material Noise Suppression in Ultrasonic Nondestructive Testing

Abstract

Ultrasonic nondestructive inspection of materials is often limited by the presence of backscattered echoes from the material structure, known as material or grain noise. The material noise can be difficult to distinguish from flaw echoes because their spectra overlap to a large extent. Due to the overlapping, application of conventional linear filters is generally not adequate for attenuating this type of noise. However, a suitable choice of the inspection frequency will cause the grain response to be considerable more noncoherent than the flaw response. This property arise from the fact that the material noise can be considered as an interference pattern made up of unresolved scatterers, while the flaw echoes will bear more resemblance to specular reflections. The implication of this difference is that the material noise can be suppressed by means of frequency diversity techniques which take advantage of its noncoherent nature. The desired frequency diverse signals may be obtained either by transmitting several pulses at slightly different frequencies or by using signal processing to create a set of signals from a single transmitted (broadband) pulse. The class of processing techniques used in the latter case will be referred to as Quasi Frequency Diversity (QFD) algorithms. The most renowned QFD algorithm is the Split Spectrum Processing (SSP) [1,2] which was developed about a decade ago. Through experimental verification the possibilities and limitations of the SSP have become evident. In order to avoid the drawbacks of the SSP we have developed a novel algorithm, which we refer to as Cut Spectrum Processing (CSP). In this paper the CSP algorithm is introduced and its operation is illustrated by an example.

Performance Evaluation of Neural Networks in Concrete Condition Assessment

“…Neural nets have been successfully applied for flaw detection of noisy titanium alloys and in flaw signature estimation (3,4). A radial basis function (RBF) network was implemented to size defects in ultrasonically tested aluminum blocks (5).…”