An automated system for characterizing ultrasonic transducers usingpattern recognition

The system consists of a 3D positioning mechanism, a motion controller, a pulser/receiver with gated-peak detector, a digitizing oscilloscope, a spectrum analyzer, and a host computer. Pattern recognition techniques were used to classify and reduce the dimensionality of the transducers. It was found that the K-means algorithm was the most successful algorithm for classifying the transducers, whereas the Baye's decision rule gave the worst performance. Feature reduction was found to be successful through the K-L transformation algorithm. Details of hardware and software implementation as well as the pattern recognition characterizing techniques and results obtained are presented. The characterizing techniques are compared