Intelligent processing of ultrasonic signals for quantitative materials testing

Some of the approaches that have been used to apply analogous, intelligent, neural-like signal processing procedures to solve a number of passive acoustic emission and active ultrasonic inverse problems are reviewed. Characteristic of these approaches is the use of a set of learning signals to develop a memory which can subsequently be utilized to process test signals to optimally recover information related to the source or the medium. In an approach suitable for solving linear problems, the memory is used to map ultrasonic source to waveform data and vice versa using an auto-associative processor. Other approaches have been developed for solving nonlinear problems. Included are multi-layered feedforward neural networks or, alternatively, a multi-dimensional regression approach called an automatic modeler. Examples of ultrasonic measurements using these approaches are summarized and the challenges for the future indicated.