Date of Award
Doctor of Philosophy
Engineering Science and Mechanics
Lester W. Schmerr
In the ultrasonic testing of materials for flaws, it is important to be able both to identify the flaw type (flaw classification) as well as to predict the flaw shape, orientation and size parameters (flaw sizing). In this work, we describe new techniques for both classification and sizing. In particular, a flaw classification technique is considered that employs mode-converted diffracted signals in a quasi-pulse-echo configuration to distinguish smooth vs. sharp-edged flaws. For the flaw sizing applications, three approaches are presented. These approaches include (1) a non-iterative equivalent sizing method, where the best equivalent ellipsoid (for volumetric flaws) or ellipse (for cracks) is found that matches the scattering data, (2) a spherical harmonics expansion algorithm and (3) the use of neural networks for equivalent flaw sizing. In addition, we consider the effect of classification information on the sizing problem, describe a technique for correcting systematic errors in sizing cracks due to the finite bandwidth of ultrasonic transducers, and present an enhanced adaptive learning method that can speed up the training of a neural network.
Digital Repository @ Iowa State University, http://lib.dr.iastate.edu/
Chiou, Chien-Ping, "Model-based ultrasonic flaw classification and sizing " (1990). Retrospective Theses and Dissertations. 9428.