Multistage adaptive noise cancellation and multi-dimensional signal processing for ultrasonic nondestructive evaluation

Ultrasonic signal processing presents several challenges with respect to both noise removal and interpretation. The interference of unwanted reflections from material grain structure can render the data extremely noisy and mask the detection of small flaws. It is therefore imperative to separate the flaw reflections from grain noise. The interpretation or classification of ultrasonic signals in general is relatively difficult due to the complexity of the physical process and similarity of signals from various classes of reflectors;Adaptive noise cancellation techniques are ideally suited for reducing spatially varying noise due to the grain structure of material in ultrasonic nondestructive evaluation. In this research, a multi-stage adaptive noise cancellation (MANC) scheme is proposed for reducing spatially varying grain noise and enhancing flaw detection in ultrasonic signals. The overall scheme is based on the use of an adaptive least mean square error (LMSE) filter with primary and reference signals derived from two adjacent positions of the transducers. Since grain noise is generally uncorrelated, in contrast to the correlated flaw echoes, adaptive filtering algorithms exploit the correlation properties of signals in a C-scan image to enhance the signal-to-noise ratio (SNR) of the output signal;A neural network-based signal classification system is proposed for the interpretation of ultrasonic signals obtained from inspection of welds, where signals have to be classified as resulting from porosity, slag, lack of fusion, or cracks in the weld region. Standard techniques rely on differences in individual A-scans to classify the signals. This thesis investigates the need for investigating signal features that incorporate the effects of beam spread and echo dynamics. Such effects call for data interpretation schemes that include a neighborhood of A-scans carrying information about a reflector. Several ultrasonic signal features based on the information in a two-dimensional array of ultrasonic waveforms, ranging from the estimation of statistical characteristics of signals to two and three-dimensional transform-based methods, are evaluated. A two-dimensional scan of ultrasonic testing is also represented in the form of images (B- and B\u27-scans). Multidimensional signal and image-processing algorithms are used to analyze the images. Two and three-dimensional Fourier transforms are applied to ultrasonic data that are inherently three-dimensional in nature (2 spatial and 1 time). A variety of transform-based features are then utilized for obtaining the final classification