ANALYSIS OF LOW-ORDER AUTOREGRESSIVE MODELS FOR ULTRASONIC GRAIN SIGNAL CHARACTERIZATION

In testing materials nondestructively with ultrasound, the grain scattering signal provides information that may be correlated to regional microstructure variation. In the Rayleigh scattering domain, grain scattering results in an upward shift in the expected frequency of broadband echoes, while attenuation caused by scattering and absorption affects the shift in a downward direction. Both upward and downward shifts are known to be dependent on grain size distribution. In this paper, the second- and third-order autoregressive (AR) models are used to evaluate the spectral shift in grain signals by utilizing features such as resonating frequency, maximum energy frequency or AR coefficients. Then, Euclidean distance, based on these features, is applied to classify grain scattering characteristics. Using both computer simulated data and experimental results, the probability of correct classification is found to be about 75% for the second-order AR model and 88% for the third order AR model, when the conditions are such that the expected shift between the center frequency of echoes is less than 4%.