INTELLIGENT PROCESSING OF ULTRASONIC SIGNALS FOR PROCESS-CONTROL APPLICATIONS

We review here some of the approaches that have been used to apply int elligent, neural-like signal processing procedures to solve a number o f acoustic emission (AE) and active ultrasonic (UT) process control me asurement problems which can be expected to have important process con trol applications. Characteristic of these approaches is the use of a set of learning signals from an array of sensors to develop a memory c ontaining prototypical pattern vectors composed of acoustic data and p rocess characteristics. This memory can subsequently be utilized to pr ocess signals to optimally recover parameters of the manufacturing pro cess. Approaches applicable for linear and non-linear problems have be en developed. For the latter, algorithms implementing a multi-layered, feed-forward neural network or alternatively, a non-parametric, multi -dimensional regression approach called an automatic modeler have been developed. Here we illustrate this modeler by its application to char acterize a drilling process and to recognize the finish of surfaces fr om the signals generated using a tactile sensor.