Detection of ultrasonic anomaly signals using wavelet decomposition

A major limitation facing the ultrasonic evaluation of materials is the high level of background noise from unresolvable grain boundaries that often mask the reflections from the target of interest in the measured signal. The split spectrum processing (SSP) technique, which is based on frequency diversity concepts, has been established as an effective method of achieving anomaly enhancement and grain noise suppression. An alternate decomposition which promises improved resolution capabilities at high frequencies for the purpose of detecting closely spaced multiple targets was presented as a natural extension to conventional SSP. In this work, wavelet decomposition and reconstruction algorithms are used to achieve a constant-Q decomposition of the signal. In recent years, wavelet techniques have emerged as useful tools in signal analysis because of their time-frequency localization properties. Two implementations based on the wavelet transform are presented here: direct implementation, which is similar to the split spectrum processing filter bank realization; and the discrete wavelet transform (DWT) which is implemented using computationally efficient pyramidial structures. Nonlinear algorithms are used to obtain the output signal from the reconstructed signals. Preliminary results indicate that these methods are quite successful in the detection of single targets, but not as effective as split spectrum processing in the resolution of closely spaced multiple targets.